Science and Technology Committee (Commons)
Wednesday 4th November 2026
Skip to latest contribution
This debate has concluded
(began 3 days, 6 hours ago)
Watch LiveSkip to latest contribution
This debate has concluded
09:31
Speaker 1
-
Speaker 1
Order, order and welcome to the Science and Technology Committee's second evidence session in our inquiry into the exciting area of low energy computing. This inquiry grew out of our Under the Microscope initiative, through which researchers drew our attention to the potential of neuromorphic computing and silicon photonics. And in our first session, we explored the science behind these technologies and their potential to make computing more energy efficient and the implications of that. But today we are going to turn to their practical application and the industry perspective, and we want to understand where these technologies could deliver meaningful benefits, how soon they might be adopted more widely, and what barriers stand in their way, in their way.
And to help us understand that. I'm really pleased to welcome our first panel, and I'm going to ask each of these industry experts to introduce themselves and their organisation. And as they explain the neuromorphic technology that they use. And so if I could start with Doctor Mangoma, could you introduce yourself, explain the technology you're using and what's so special about it?
And to help us understand that. I'm really pleased to welcome our first panel, and I'm going to ask each of these industry experts to introduce themselves and their organisation. And as they explain the neuromorphic technology that they use. And so if I could start with Doctor Mangoma, could you introduce yourself, explain the technology you're using and what's so special about it?
09:33
Speaker 2
-
Speaker 2
Brilliant. Thank you, chair, and thank you all for having us. My name is Tanya Mangoma. I am the Chief Product officer at Analogue Insight, where I design and IP house, which basically means for companies who take the ideas from a power point presentation all the way to silicon. So we work with all the big foundries to be able to deliver this and design for them. We focus on a specific skill set called analog and mixed signal design. And this is a key component for neuromorphic computing, especially when you're talking about analog neuromorphic computing. For myself, I have a PhD in neuromorphic computing, and since graduating, I've been trying to commercialize the technology within the UK, facing many barriers, but overcoming some.
And hopefully today I'll be able to share more about what's happening in the ecosystem and how we could help as analogue, but also as personalities.
And hopefully today I'll be able to share more about what's happening in the ecosystem and how we could help as analogue, but also as personalities.
09:33
09:33
Speaker 2
-
Speaker 2
In computing? There are two types digital, which is basically on and off. Yeah. And then analog, which is basically signal processing, which is the analog part. So for signal processing, this was done years ago. And actually spinnaker was one of the examples they used digital. So spinnaker I don't know if you're aware of Steve , what he did in Manchester with spinnaker. He made digital compute look like analog signal because he wanted to replicate a spiking neural networks in in compute. Right. But what I actually do, what I do for my PhD, was actually use technologies and chips and materials that actually work as spiking neural networks inherently.
So you don't have to emulate them. You actually have, you actually emulate the brain. You don't have to simulate what they do in the software.
So you don't have to emulate them. You actually have, you actually emulate the brain. You don't have to simulate what they do in the software.
09:34
09:34
09:34
09:34
Speaker 2
-
Speaker 2
So there is neural networks as we know them at the moment. Then there's the spiking neural network that brings in a time component. So then all of a sudden you have this event based kind of like compute that you can actually harness. No longer do you train something and then keep it trained. So at the moment you kind of like train an LLM and train GP, and then you have to wait for open AI to train another model before it's released, right? Spiking neural networks, basically it trains as you actually do work in real time.
So that's why you get spiking plasticity or spiking neural network coming in.
So that's why you get spiking plasticity or spiking neural network coming in.
09:35
09:35
Speaker 3
-
Speaker 3
Thank you. So I'm Tony Kenyon and I have three hats on. I suppose . One is I'm a professor in electronic electrical engineering department at UCL. Second one is that I also have formed a spin out company to commercialize novel neuromorphic Adjacent memories. And the third one is I'm director of a National center, an IC and innovation and knowledge center called Neuro Air, which is looking to translate technologies in neuromorphic computing hardware from the lab through to early stage demonstrators and all the way through, ultimately to commercialization, to real world impact. And that's a center that encompasses nine universities, plus the National Physical Laboratory.
We have somewhere between 30 and 40 industry partners of various types. We have also an extensive international network as well, and we've been working since October 2025. So we're just over a year old now, and we have another four years to run.
We have somewhere between 30 and 40 industry partners of various types. We have also an extensive international network as well, and we've been working since October 2025. So we're just over a year old now, and we have another four years to run.
09:36
Speaker 1
-
Speaker 1
Great, great. So so we've identified some of the implications. I want to go into that a little bit deeper, but can you just for our benefit, is it true to say that neuromorphic computing is looking to emulate more closely how the brain works than digital computing? Would you agree with that as a sort of a very high level summary? Yes, yes.
09:37
09:37
Speaker 3
-
Speaker 3
Yes. Partly true. So so yes, there are elements that we would like to emulate, but there are also lessons that we want to take from the brain that we can apply in different ways. So an example would be when we look at the brain and we look at, for example, the speed at which the brain does the compute, and that's on a sort of biological time scale. But if we interface that with, for example, photonics, where we have much higher data rates, and then we can start to work in, in different applications, right? Yes, very high speeds and taking inspiration and an example that often hear people talk about is if we think about flying.
So when we built a flying machine, so jets don't flap their wings, but we've taken inspiration from birds. So that's the sort of idea that we emulate some things, but then we want to go further.
So when we built a flying machine, so jets don't flap their wings, but we've taken inspiration from birds. So that's the sort of idea that we emulate some things, but then we want to go further.
09:37
09:38
09:38
09:38
Speaker 2
-
Speaker 2
Okay, I'll speak up a bit. Is this better? Okay, great. So when it comes to neuromorphic computing, the big area that is noticeable is the energy aspect of it, right? So as soon as you have neuromorphic or brain like computing, you automatically have lower energy consumption when you're doing AI inference and training in general. So that is well established and well knowledge. So that is the first part of what we could actually benefit from having neuromorphic computing. The second part is actually when it comes to an area called sense detect Act or edge computing, where basically you have areas such as autonomous vehicles, UAVs, and kind of like drone capacity and cybersecurity, where you want to be able to sense something extremely quickly and act extremely quickly.
That is where neuromorphic computing actually shines a lot more than our conventional computing. So those two areas on the energy aspect and what it can actually do for edge computing and area now called physical AI in general. Those are the two areas that I see the most benefit in. I don't know if.
That is where neuromorphic computing actually shines a lot more than our conventional computing. So those two areas on the energy aspect and what it can actually do for edge computing and area now called physical AI in general. Those are the two areas that I see the most benefit in. I don't know if.
09:39
09:39
Speaker 3
-
Speaker 3
To add anything? I can maybe add something to that. So where where there are other opportunities as well. So absolutely in energy and security. But the third one is we, we talk a lot about data centers, about the fact that we have to build bigger and bigger and faster and faster data centers to deal with the deluge of data that we're experiencing at the moment. But I think neuromorphic computing offers perhaps a different way of doing things, which is maybe we don't need to generate the data in the first place. So here I make a distinction between a signal.
So something in the environment that we're detecting data, which is what we turn the signal into conventionally, and then information, which is the important thing. And in fact, when we look at the way the brain operates, the brain doesn't take, for example, the whole environment and commit that to memory. Or what it does is it pays attention to those things that are important. Now, if I give an example, maybe a concrete example is a technology known as an event camera. So in an event camera there, what we have is a series of pixels in the camera that only produce a signal or produce data when something changes in the environment.
So what that means is that if something changes in a small part of the environment, that's the that's the data that's produced rather than digitizing the whole image. Yeah. And if we can take that example and apply that in other domains as well as vision, then you can see how the potentially huge advantages where now we're not having to produce the data in the first place. And I think maybe sorry.
So something in the environment that we're detecting data, which is what we turn the signal into conventionally, and then information, which is the important thing. And in fact, when we look at the way the brain operates, the brain doesn't take, for example, the whole environment and commit that to memory. Or what it does is it pays attention to those things that are important. Now, if I give an example, maybe a concrete example is a technology known as an event camera. So in an event camera there, what we have is a series of pixels in the camera that only produce a signal or produce data when something changes in the environment.
So what that means is that if something changes in a small part of the environment, that's the that's the data that's produced rather than digitizing the whole image. Yeah. And if we can take that example and apply that in other domains as well as vision, then you can see how the potentially huge advantages where now we're not having to produce the data in the first place. And I think maybe sorry.
09:40
Speaker 1
-
Speaker 1
Yeah. I mean, I think that does a great I will we'll come in more detail to examining in more detail. And before, before I hand over to, to me, I just want to actually, I should possibly declare, declare an interest as an electrical engineer and a fellow of the Institution of Engineering Technology. It's one of the reasons why I find this discussion so interesting. And the points you make about that, it's much less signal. It's much less data to, to, to share the changes rather than to, to share to or to generate everything. But, but, but on, finally, on sort of the application that sort of jumps out and that Doctor Montgomery mentioned about reducing the power requirement.
The what I've got here is that the human brain uses 20W of power to process process data that would take around one megawatts of power within a data center. So that is to say that the human brain , 2% , the power usage of existing compute. Surely that would that would revolutionize and have a have a real impact on the the value of the existing data center infrastructure investments, which are in the trillions of pounds, trillions of dollars and trillions of pounds.
Is that true?
The what I've got here is that the human brain uses 20W of power to process process data that would take around one megawatts of power within a data center. So that is to say that the human brain , 2% , the power usage of existing compute. Surely that would that would revolutionize and have a have a real impact on the the value of the existing data center infrastructure investments, which are in the trillions of pounds, trillions of dollars and trillions of pounds.
Is that true?
09:42
Speaker 2
-
Speaker 2
Yeah. No, absolutely. That's true. And I think to to really clarify why that is so current data centers are built on digital compute. And for that it means there is a compute aspect on one end, the memory on one end, and data is moving between them and the memory is always on. So the memory is always consuming power, the data is always moving, and if it's moving from one server to another, or one data facility to another, or from one compute edge to another. That is all consuming energy. That's where all the energy is going.
It's not going into compute. It's going to actually just moving the data and the storage of the memory that is volatile. So when it comes to the brain, it does not separate memory and compute. They one thing, they exist in one place. So all of a sudden you're not moving data. And when you're not moving data, all of a sudden that energy cost disappears in all honesty. So even applying the simplest terms of neuromorphic computing, which is kind of like co-location of memory and, and, and compute, you actually get the, the added advantages of not moving that data in the first place, right?
And then the second aspect then comes with things like spiking neural networks, where the spike in the timing, the timing dependency of that spike goes to what you were talking about, about signal processing, which signal is important, what information should be moved, what data should be generated. And then that's where that comes in of cutting even further, what data is moved between those two boundaries.
It's not going into compute. It's going to actually just moving the data and the storage of the memory that is volatile. So when it comes to the brain, it does not separate memory and compute. They one thing, they exist in one place. So all of a sudden you're not moving data. And when you're not moving data, all of a sudden that energy cost disappears in all honesty. So even applying the simplest terms of neuromorphic computing, which is kind of like co-location of memory and, and, and compute, you actually get the, the added advantages of not moving that data in the first place, right?
And then the second aspect then comes with things like spiking neural networks, where the spike in the timing, the timing dependency of that spike goes to what you were talking about, about signal processing, which signal is important, what information should be moved, what data should be generated. And then that's where that comes in of cutting even further, what data is moved between those two boundaries.
09:43
Speaker 1
-
Speaker 1
So that's fascinating. And and it also speaks to the, the investment in RAM, which, which the AI companies are making. And, you know, Elon Musk was, was it I was sorry, Sam. Sam Altman who was saying that there shouldn't we shouldn't be having RAM in our computers because it was so expensive and so and so valuable that it should be saved for data centers. This is a whole new way of doing things, which would mean that, you know, that that, that, that we wouldn't, that that trade off or that demand wouldn't, wouldn't, wouldn't make, wouldn't be needed anymore.
09:43
Speaker 2
-
Speaker 2
Indeed. And I would actually like to highlight. So the RAM technology that is used at the moment, which is Dram, if you look at South Korea and the economy and how much it's been propped by those technologies, SK Hynix, Samsung and what they've managed to do. And then you look at that technology when it comes to reram, for example, and non-volatile memory technologies in UK universities, IP that has actually been secured and how much we are not doing about it, it is quite saddening.
09:44
09:44
09:44
09:44
09:44
09:44
09:44
09:44
Speaker 4
-
Speaker 4
Thank you. And thank you for explaining that so clearly. And it's fascinating and very exciting. So I guess my question is about the bigger picture then. So you've talked to us about the technical possibilities of. It seems very clear. What are you finding in terms of, I guess, the political sort of reception to this or kind of where the barriers are? What are the sort of sense of we've got? I think we've had a bit of a sense from you in sort of the initial notes of kind of what the, you know, basically the IP is there and that it needs to focus.
What do you what's the reception of it? And we talked about data centers in terms of the current ecosystems, an awful lot of talk about AI. Is there resistance to this because it's something different and new, you know, is there is there an appetite, but it's just not getting enough funding? What's what's the kind of the landscape looking like in terms of building this?
What do you what's the reception of it? And we talked about data centers in terms of the current ecosystems, an awful lot of talk about AI. Is there resistance to this because it's something different and new, you know, is there is there an appetite, but it's just not getting enough funding? What's what's the kind of the landscape looking like in terms of building this?
09:45
09:45
09:45
Speaker 2
-
Speaker 2
So there are two areas actually, especially when you talk about the UK. So the UK can do a lot of the material science research. And that's where we hold majority of the IP. But to commercialize that IP you need to go to a foundry service. And the foundries have specific recipes that they will allow you to use. We don't have access to those. So if you wanted to commercialize anything you do within research
09:47
Speaker 3
-
Speaker 3
perception was the UK doesn't do semiconductors. Now that's changed. And the fact that we're having this discussion, I think is an example of that. And I think it's a combination of both public and private money. So I think there needs to be an ecosystem that allows creates support for early stage companies to get over this, this, this valley of death, which is, you know, that that bit between what you can do in a university and what you need to then go to a foundry to, to unlock the capabilities of TSMC or UMC. And first of all, to say that if you're developing a new technology in this space, then a rule of thumb is that it will take you $100 million to get from having the technology in the laboratory to having something that is commercial now, that can't all come from public money.
So that needs to come from investment. It needs to be patient capital. And in the UK, what I think what's holding us back is the availability of, first of all, the availability of sufficient capital. But the second thing is the risk appetite as well. So it's it's very true that if you go to the West Coast, if you go to California and you present the same technology, then you're likely to get an order of magnitude more funding and a much greater risk appetite. And the way I sort of characterize this is in the UK, if you set up a company and you attract some investment and then the company fails, then that's shame on you, shame on your family.
For generations in California, if you do the same thing, then it's okay. You failed at that. You've learned something. What can we give you now? Because we believe in the team, not necessarily the technology, but we think the team can can develop another technology. So that risk appetite is is very different. And then around that you need a broad ecosystem. Tanya spoke about prototyping and pilot lines. You absolutely need that and you need test facilities. You need access to design tools and so on and so forth where it would be something the government could could do very materially to help with.
This would be around procurement, for example. So where a new technology is developed and is promising, for example, for defense or healthcare or whichever sector, robotics, then government could step in and say, that will be we will be the first customer for for this new technology.
So that needs to come from investment. It needs to be patient capital. And in the UK, what I think what's holding us back is the availability of, first of all, the availability of sufficient capital. But the second thing is the risk appetite as well. So it's it's very true that if you go to the West Coast, if you go to California and you present the same technology, then you're likely to get an order of magnitude more funding and a much greater risk appetite. And the way I sort of characterize this is in the UK, if you set up a company and you attract some investment and then the company fails, then that's shame on you, shame on your family.
For generations in California, if you do the same thing, then it's okay. You failed at that. You've learned something. What can we give you now? Because we believe in the team, not necessarily the technology, but we think the team can can develop another technology. So that risk appetite is is very different. And then around that you need a broad ecosystem. Tanya spoke about prototyping and pilot lines. You absolutely need that and you need test facilities. You need access to design tools and so on and so forth where it would be something the government could could do very materially to help with.
This would be around procurement, for example. So where a new technology is developed and is promising, for example, for defense or healthcare or whichever sector, robotics, then government could step in and say, that will be we will be the first customer for for this new technology.
09:49
Speaker 4
-
Speaker 4
And then I guess just the last sort of two part question is kind of, do you think that this space is unique in some of the stuff that you've described? Because obviously we hear those kind of concerns for all different technologies, and there's lots of different researchers and different fields who would all say, you know, government needs to support this one. So I guess a, do you feel that this area of research is particularly disadvantaged for various reasons and sort of either which way or particularly if it is, what's your kind of pitch, I guess, to government or to private sector as to why this technology, what's the opportunity with the technology, I guess, particularly around sort of AI and sort of retrofitting and what, you know, in terms of what's existing there compared to some of the other things that we'll be pitching for similar kind of investment?
09:50
Speaker 3
-
Speaker 3
No, it's not unique. So a lot of the things that we're talking about are common to all semiconductor industry. There are particular things around neuromorphic, and I'm sure you'll hear about particular issues around photonics as well. One of those being that it's a very interdisciplinary or multidisciplinary field. And there's a lot of excellent work, particularly in the UK that's a little bit disparate. So bringing all of that together I think is really important. The second thing is that a lot of this is about developing a new technology in the sense of working potentially with new materials, and that's very difficult to do in silicon based semiconductor industry.
So there's a lot of lot of work that needs to be needs to be done there. Sorry. What was the second part of your question.
So there's a lot of lot of work that needs to be needs to be done there. Sorry. What was the second part of your question.
09:51
Speaker 4
-
Speaker 4
In terms of what I guess what therefore, what is the pitch to kind of in layman's terms to government and to private industries, to why, why should we be focusing on this when there's obviously lots of focus on just AI in general, or data centers or all the different other researchers? What, why, why do you think this is the thing that one of the things that the UK should really.
09:51
Speaker 3
-
Speaker 3
Focus on, I think I'd probably say four things there. So one is we've mentioned the potential benefit in terms of power consumption and making things like data centers, but also edge compute. So think of smart sensors, think of wearables and so on. The second one is security that we mentioned. The third one is being able to do things differently as we talked about. And, and then I think the other thing to mention is we're talking about this as research. Actually, it's a technology that exists. So the first neuromorphic technology was rolled out in the 1990s, and it was a chip that was used in Trackballs.
So actually tracking the position of the trackball, that was a neuromorphic chip. There are now event cameras that I mentioned that are deployed on the International Space Station. They're also deployed in ground stations to monitor satellites, and there are commercial chips that are available. So it's the technology that's here that needs support to grow further to really fulfil its potential.
So actually tracking the position of the trackball, that was a neuromorphic chip. There are now event cameras that I mentioned that are deployed on the International Space Station. They're also deployed in ground stations to monitor satellites, and there are commercial chips that are available. So it's the technology that's here that needs support to grow further to really fulfil its potential.
09:52
09:52
Speaker 2
-
Speaker 2
If I could actually add on to that, I wouldn't say we have particularly different problems from semiconductor areas, but I would say like, we have so much strength in the UK, we are not doing our job, if that makes sense. We have spinnaker, which was made by Steve Furber. Professor Steve is like well decorated. It was the first neuromorphic computer that did 1% of a mouse brain, which sounds like nothing, but that is groundbreaking, right? It's a full system that does the software and the hardware. It interfaces with everything. At the end of his book, he has a chapter of what could this be used for?
One of the things was robotics. And if you understand anything about AI and AI computing, where we are going in AI and the noise that is being made by Nvidia and everyone else, fiscal AI is the next thing, and robotics is the center. We already have that technology developed here in the UK, and it was UK money spent. Could we have a smaller version of spinnaker that is made in the UK, manufactured by the UK? Of course we could use pragmatic, a smaller foundry that has flexible semiconductors. We could use that. Right. And could we have novel material scientists making sure that everything works, their analog computer engineers making sure that everything works.
Of course we can do that. We have all the missing components that are needed for us to capture the next wave for the next frontier compute, but we're just not using it. So no, we do not need something super different. But our pitch is everything is there. We just need coordination and maybe a little bit of money to make it happen. But it's already there and we actually paid for it already.
One of the things was robotics. And if you understand anything about AI and AI computing, where we are going in AI and the noise that is being made by Nvidia and everyone else, fiscal AI is the next thing, and robotics is the center. We already have that technology developed here in the UK, and it was UK money spent. Could we have a smaller version of spinnaker that is made in the UK, manufactured by the UK? Of course we could use pragmatic, a smaller foundry that has flexible semiconductors. We could use that. Right. And could we have novel material scientists making sure that everything works, their analog computer engineers making sure that everything works.
Of course we can do that. We have all the missing components that are needed for us to capture the next wave for the next frontier compute, but we're just not using it. So no, we do not need something super different. But our pitch is everything is there. We just need coordination and maybe a little bit of money to make it happen. But it's already there and we actually paid for it already.
09:53
09:53
09:53
09:54
09:54
09:54
Speaker 2
-
Speaker 2
Okay. So all the chips that we make that are scalable or mass manufactured come from foundries. They're called foundries. These are the TSMC of the world, the global foundries of the world, TSMC, Taiwan. So basically, when everyone is talking about how Taiwan is important for computer chips and everyone's going to fight for them.
09:54
09:54
09:54
09:54
Speaker 2
-
Speaker 2
So they're they're like a handful of companies who can actually mass manufacture chips for everything that we do from defense telephones, anything that you can think of like to AI chips. And we don't have that capability in the UK. We should not try to emulate that capability. But then those companies are the only people who can take the research that we do in our labs and put it into their systems to say, okay, for the the next 90% of chips that are going to be made, we're going to be using IP from Cambridge, from neuromorphic computing, because we have actually made it happen.
But to get to that door of global foundries working with us, TSMC, working with us, it can't be me. Because who am I? Right? It has to be coming from government.
But to get to that door of global foundries working with us, TSMC, working with us, it can't be me. Because who am I? Right? It has to be coming from government.
09:55
09:55
09:55
09:55
Speaker 7
-
Speaker 7
And just help me understand how the technology path works here, because from my understanding that this neuromorphic computation, it's effectively functional filtering creates sparse data sets. I thought that needed a different architecture up and down the whole system. But it sounds like you think this could be implemented into mainstream applications quite quickly. So just explain that distinction for me, if that's okay.
09:55
Speaker 2
-
Speaker 2
Absolutely. So I'll take actually spinnaker as an example, right? When we think about application data sets, we think about LLMs and how they are done. Currently, when we think of things like ChatGPT and kind of like matrix multiplication. But then here we're doing it ever so slightly different and we're training it faster and doing event based and the stack is already developed. That's why I was bringing up that. When we look at Nvidia, they have already like, you know, the software part of it and also the hardware that's already done and everything is implemented. The University of Manchester already made the full stack for, for, for, for.
And then they, they also trade the training with conventional computing versus the training with neural networks. And how much more better and faster it is. So all of this is already done. So the stack and the compatibility is there for use cases where event based kind of sensing or event things are needed.
And then they, they also trade the training with conventional computing versus the training with neural networks. And how much more better and faster it is. So all of this is already done. So the stack and the compatibility is there for use cases where event based kind of sensing or event things are needed.
09:56
09:56
09:56
09:56
09:56
09:56
Speaker 8
-
Speaker 8
Low energy computing. We're a high energy committee, so we want to find ways in which we can help elicit what's needed. Now, following on from my colleague's comments, does the UK actually have the industrial capacity to do the manufacturing at commercial scale? And what would that look like? Explain it to me.
09:57
Speaker 3
-
Speaker 3
Right. So it's it's a very complex system in semiconductors. So I don't think there is any country with possible exception of Taiwan, that has the capability to the full stack of fabrication at the most advanced technology. However, there are opportunities at. I think what you need to do is you need to build partnerships. And this is how it works in semiconductors, that you might have one country that is very strong in design and the UK is very strong in that space. Another country could be very strong in the fabrication of chips, and that might be Taiwan packaging is something else.
And then actually putting that into systems is something else again. So it's a very globalized industry, and I don't think that's going to change any time soon. So there's a lot of discussion around this concept of sovereignty. What does sovereignty mean. And it's a very difficult thing to pin down. But I think in this space, the conversations are going along the lines of, well, what we really want is we want to make sure that we have ownership of some part of that chain. And an example that I can give on that is the Dutch company ASML.
So they manufacture the tools that are used to create the to write the structures, the very small structures on the nanometer scale that are in the most advanced chips. They're the only company in the world that can do this. It's fantastically expensive to to develop this technology. And it's taken 20 plus years to develop it. Nobody else can do it. So they own that part of the supply chain. Now, is there something we could do.
And then actually putting that into systems is something else again. So it's a very globalized industry, and I don't think that's going to change any time soon. So there's a lot of discussion around this concept of sovereignty. What does sovereignty mean. And it's a very difficult thing to pin down. But I think in this space, the conversations are going along the lines of, well, what we really want is we want to make sure that we have ownership of some part of that chain. And an example that I can give on that is the Dutch company ASML.
So they manufacture the tools that are used to create the to write the structures, the very small structures on the nanometer scale that are in the most advanced chips. They're the only company in the world that can do this. It's fantastically expensive to to develop this technology. And it's taken 20 plus years to develop it. Nobody else can do it. So they own that part of the supply chain. Now, is there something we could do.
09:58
09:58
09:58
09:58
09:58
09:58
09:59
09:59
Speaker 3
-
Speaker 3
Right. So that was also a very complicated process. So it involved Phillips Phillips, who were obviously a Dutch company who were German, but an awful lot of investment from people like Intel and US companies, US government, Europe as well, over a period of about 20 years. At the beginning, it wasn't actually known whether it was doable, whether actually to create a machine. And this is now, I think, the most complex machine that has ever been, ever been designed and manufactured, by the way, to buy one will cost you about $500 million to, to write these very fine structures.
3.9 million 500 million, 500 million.
3.9 million 500 million, 500 million.
09:59
09:59
09:59
09:59
10:00
Speaker 8
-
Speaker 8
So you've explained the process by which that happened, but you hinted in your first answer to my question that we needed to identify a part. This is sounds very similar to the quantum discussion to me, but we need to identify a specific thing which we which we are good at. Yes. Where are we on that?
10:00
Speaker 2
-
Speaker 2
Yeah. So I would actually emphasize three areas the UK are really, really good at when it comes to capabilities. Obviously there's going to be photonics. There is also neuromorphic, but there's also silicon carbide which is down in well, so these are the. silicon silicon carbide. So power electronics, silicon carbide which is based in Wales is a big cluster. And across all three of them we are very lucky to have smaller foundries. So we can't do the high scale technologies that need us to have conversations with Taiwan to be able to manufacture, but for neuromorphic computing, for small chips that can go into, for example, drones, we could use pragmatic because we can actually manufacture and have that sovereignty here.
That's the ownership. We have spinnaker. That is the top to bottom software and hardware developed already. We have capability to manufacture here in Durham in pragmatic. That is something that we can own and actually deploy in the UK. So that is sovereignty right there. Then the similar applies to kind of like Cornerstone and photonics. The similar applies to silicon carbide in Wales. Those are the three clusters we have that have everything manufacturing deployment. Actually the people who know the technology and have the experience, plus also the materials research that actually backs all of this. So that is something highly defensible.
That's the ownership. We have spinnaker. That is the top to bottom software and hardware developed already. We have capability to manufacture here in Durham in pragmatic. That is something that we can own and actually deploy in the UK. So that is sovereignty right there. Then the similar applies to kind of like Cornerstone and photonics. The similar applies to silicon carbide in Wales. Those are the three clusters we have that have everything manufacturing deployment. Actually the people who know the technology and have the experience, plus also the materials research that actually backs all of this. So that is something highly defensible.
10:01
10:01
10:01
10:01
Speaker 8
-
Speaker 8
When this committee did its report, which we called flying blind, we identified a well known problem. I'm the MP for Cambridge, and it's a regularly discussed about how we take things, scale them up and our inability to do that. How significant is that challenge in neuromorphic computing? Massive. Good answer.
10:02
Speaker 3
-
Speaker 3
It's it's probably the biggest challenge, I would say, because I think there are a lot of things that are associated with that, which include the building of an ecosystem, bringing together a disparate community. And then all of the things that I talked about earlier, providing the, the capital, the patient capital and the risk appetite.
10:02
Speaker 3
-
Speaker 3
Okay, so what I'll say on this is that neural, where I see in September published a roadmap for neuromorphic computing going out to 2050, and that made a series of recommendations. So we had 26 different, different points that we recommended across five different application areas. And I have copies of that. So I can, I can leave copies of digest of that with you if you'd like to read it, but that has a number of recommendations for government, which are around the obviously investment but sustained investment. So making sure that. The for example, Norway has a five year term , extending that thinking about having a national neuromorphic strategy.
So we have strategies in semiconductors. We have strategies around quantum and other areas. Neuromorphic, which is a technology that exists, would benefit from having a strategy looking at commissioning a feasibility study on on shared prototyping facilities. And this would cut across as well as neuromorphic other areas in semiconductors. And I'm sure you're hearing the same things from from other people in that area, funding national testbeds. As we generate these new as we develop these new technologies, we need to have testbeds to, to really prove their mettle, prove, prove, prove their application. I mentioned government potentially acting as an early customer through public procurement funding.
Going to your question about are we bringing people in from outside the UK funding trainings and, for example, funding centres for doctoral training in this space? And then, for example, including neuromorphic computing and discussions of the AI Energy Council, because we've talked about the benefits to in terms of.
So we have strategies in semiconductors. We have strategies around quantum and other areas. Neuromorphic, which is a technology that exists, would benefit from having a strategy looking at commissioning a feasibility study on on shared prototyping facilities. And this would cut across as well as neuromorphic other areas in semiconductors. And I'm sure you're hearing the same things from from other people in that area, funding national testbeds. As we generate these new as we develop these new technologies, we need to have testbeds to, to really prove their mettle, prove, prove, prove their application. I mentioned government potentially acting as an early customer through public procurement funding.
Going to your question about are we bringing people in from outside the UK funding trainings and, for example, funding centres for doctoral training in this space? And then, for example, including neuromorphic computing and discussions of the AI Energy Council, because we've talked about the benefits to in terms of.
10:04
10:04
10:04
10:04
10:04
Speaker 8
-
Speaker 8
Of things, I think it's quite a long list. We can we can push forward. Just going back to something I think Tanya was talking about a bit earlier, the opportunities of combining neuromorphic computing with silicon photonics. Could you say a little bit about that? What are the technical challenges around that?
10:04
Speaker 2
-
Speaker 2
So I would actually leave the silicon kind of like photonics and photonics work to my colleagues, unless you have something to add. So they're going to come later after this. But then just briefly on that on neuromorphic silicon photonics, it's something that in the future, using our silicon silicon photonics base in cornerstone and also us having the neuromorphic capability in the UK, this is something that we can see as a strength that we can actually harness. So if we could manage to actually strengthen both areas, which is the photonics part and neuromorphic parts, this converge, the architectures converge into silicon photonics, neuromorphic, which is something that we could really be leaders in in the UK, this is about bringing that capability and that sovereignty capability.
This is one of the areas that is really, really strong. But I'll leave my colleagues in the next panel to to discuss.
This is one of the areas that is really, really strong. But I'll leave my colleagues in the next panel to to discuss.
10:05
10:05
10:05
Speaker 3
-
Speaker 3
So thank you. Okay. So the source of devices that we work on in my team and others across across the UK work on these devices called Memristors. Memristors are devices whose electrical properties depend on their past history. Now, what we've been able to demonstrate is they can be sensitive to light, so they can react to, to, to light. They can act as storage so they can be memories, but they can also act as compute elements in their own right. They behave like neurons and synapses. So now you can imagine having something like an artificial retina or a sensor where all of the devices are on one plane.
So rather than the conventional system where you would have a sensor and then that would be connected externally to a processor, and that would be connected in turn to memory. Now you can have all of that in one device. So that saves enormously on area on power and adds extra functionality.
So rather than the conventional system where you would have a sensor and then that would be connected externally to a processor, and that would be connected in turn to memory. Now you can have all of that in one device. So that saves enormously on area on power and adds extra functionality.
10:06
10:06
10:06
Speaker 6
-
Speaker 6
Yeah. Fascinating. Sorry. I was just quite a lot of this goes way over my head. I, I have difficulty visualizing what you're actually talking about in terms of the science, but I think I broadly get it. I, I had some questions, which I think you broadly answered about the government funding situation, which is better than it was, but still could be better still. And I wanted to ask you some questions about coordination, but can I start with just a question about the kind of business model, because as you as you said, Doctor Mangoma, there are all the elements are kind of there, including the design capability, right?
Which is where we really do well. But what I don't understand in this is in other industries, like, I don't know, pop music, right? If I write a pop song, I'm basically the person that makes the money, even though someone else is pressing the vinyl, right? They license my vinyl. And yet in, in the world of compute, it feels like actually all the value is at the vinyl pressing end. And while we might have some amazing design shops like arm, right? Actually in terms of whether they're not $1 trillion company, right? Like, like TMC are right.
Whereas is that why is the value pyramid inverted? Is that something to do with IP? Is that what's the. So if we want to capture some of the value. Sorry. The reason I ask is we want. It's not realistic, I think for us to become a TMC foundry without putting, you know, a very significant proportion of our national GDP into building some kind of facility. Similarly, it'll take us 20 years to catch up with the Dutch with their little etching machine or massive etching machine. Right? So if what we do is the basic science and design, how do we how do we capture more of that value?
So we become as vital to the world economy as TMC are.
Which is where we really do well. But what I don't understand in this is in other industries, like, I don't know, pop music, right? If I write a pop song, I'm basically the person that makes the money, even though someone else is pressing the vinyl, right? They license my vinyl. And yet in, in the world of compute, it feels like actually all the value is at the vinyl pressing end. And while we might have some amazing design shops like arm, right? Actually in terms of whether they're not $1 trillion company, right? Like, like TMC are right.
Whereas is that why is the value pyramid inverted? Is that something to do with IP? Is that what's the. So if we want to capture some of the value. Sorry. The reason I ask is we want. It's not realistic, I think for us to become a TMC foundry without putting, you know, a very significant proportion of our national GDP into building some kind of facility. Similarly, it'll take us 20 years to catch up with the Dutch with their little etching machine or massive etching machine. Right? So if what we do is the basic science and design, how do we how do we capture more of that value?
So we become as vital to the world economy as TMC are.
10:08
Speaker 2
-
Speaker 2
That is a very good question. And that is where government, I think, could make the biggest impact here, because for all the things that we do in the UK, which is IP, when it comes to kind of material science and fundamental kind of like research, for it to make sense, it has to be translated into a commercial kind of like process. And that means talking to the TSMC of the world. That's the only way we can capture value there. So we can't capture value by becoming the new TSMC so that we can commercialize our IP, because that doesn't work.
The only way for us to actually capture is to put it into their system. And that's the value capture that we actually get from that aspect. And the only other way I can think of is when it comes to home, is pragmatic. I think I mentioned it a couple of times, pragmatic in paragraph. There are two foundries. There are smaller foundries. They can't do complex chips as the ones that are done in TSMC, but they will allow you to have some type of like, you know, smaller chips and prototypes and something you can sell for like edge compute that you can actually capture value much more quicker using those facilities.
So that's one end of it, right? The other end of it is then actually the design that we have like spinnaker. So spinnaker, which is the Manchester kind of software and hardware, kind of like facilities that has already been sold to Dresden. It's no longer in the UK, sorry, Dresden, Germany. So I don't know why that happened, but hopefully we can get it back because again, if we wanted to capture real value, it's about supporting work that's already been proven that it works, software and hardware and now actually saying, how can we commercialize that? Whether it comes to it being used for robotics, physical AI, or for autonomous vehicles, how could we take that architecture and spin it out?
That's how you get that value. And to give you context, the architecture that spinnaker has is so novel. It would have made the level of an Nvidia for spiking neural networks. That's how good that is. Wow. And we let it go.
The only way for us to actually capture is to put it into their system. And that's the value capture that we actually get from that aspect. And the only other way I can think of is when it comes to home, is pragmatic. I think I mentioned it a couple of times, pragmatic in paragraph. There are two foundries. There are smaller foundries. They can't do complex chips as the ones that are done in TSMC, but they will allow you to have some type of like, you know, smaller chips and prototypes and something you can sell for like edge compute that you can actually capture value much more quicker using those facilities.
So that's one end of it, right? The other end of it is then actually the design that we have like spinnaker. So spinnaker, which is the Manchester kind of software and hardware, kind of like facilities that has already been sold to Dresden. It's no longer in the UK, sorry, Dresden, Germany. So I don't know why that happened, but hopefully we can get it back because again, if we wanted to capture real value, it's about supporting work that's already been proven that it works, software and hardware and now actually saying, how can we commercialize that? Whether it comes to it being used for robotics, physical AI, or for autonomous vehicles, how could we take that architecture and spin it out?
That's how you get that value. And to give you context, the architecture that spinnaker has is so novel. It would have made the level of an Nvidia for spiking neural networks. That's how good that is. Wow. And we let it go.
10:10
10:10
10:10
10:10
10:10
10:10
Speaker 3
-
Speaker 3
Can I give a sort of not quite a counterexample, but a slightly more encouraging example, which is we think of Google DeepMind. So that was a spin out and that was developed obviously here in the UK. Now, admittedly, that was then bought by Google, but it remained in the UK. So we had inward investment, which is, I think, something we'd all agree is is a good thing. So what we need to do is, is look at that and say, why was that the case? And I think the reason that it stayed in the UK is because there was an ecosystem in London or in the UK that was attractive for that type of, of spin out of company.
And that's the research base, the proximity of the Turing, the Crick and others. So having that ecosystem to attract the inward investment, I think you can get into a sort of virtuous circle where if you have a number of examples of that, where maybe US investment or other investment comes to the UK and stays in the UK, then you build up enough critical mass that you then start to develop your own indigenous industry that stays in the UK rather than being.
And that's the research base, the proximity of the Turing, the Crick and others. So having that ecosystem to attract the inward investment, I think you can get into a sort of virtuous circle where if you have a number of examples of that, where maybe US investment or other investment comes to the UK and stays in the UK, then you build up enough critical mass that you then start to develop your own indigenous industry that stays in the UK rather than being.
10:11
Speaker 6
-
Speaker 6
Bought out. I understand that, but even Google mind just becomes a huge funnel for currency to the United States in the end. I mean, they leave some value here, but not by no means all of it. So just just going back then you seem to imply, doctor, that what's required in this sector is a kind of guiding mind. I think you said something needs coordination. Who who should who somewhere in the government architect.
Is that a Ukri thing? Is it a where is the guiding mind that should be sitting here?
Is that a Ukri thing? Is it a where is the guiding mind that should be sitting here?
10:12
Speaker 2
-
Speaker 2
So I think the UK Semiconductor Center was built exactly for this. I don't know how far they've gone down this pathway of actually implementing this, but I do know that there are people who know a lot about semiconductors. That's the first thing I'll do. I'll also add this part in there that I have spoken to, like the UK government, semiconductor sectors. And when we do ask them to implement some of these areas, like, you know, we are making materials, technologies, we do need input to be able to get into TSMC, UMC so we can actually get our technologies there.
Their response was not that positive. So for us, I think what would be useful is for this committee to recommend that they actually do implement some of the areas that we want to actually get into. If we're going to get our technologies from the lab into the Fab, we do need government intervention, and that is where government is the only person who can act. So areas like that is where we see the.
Their response was not that positive. So for us, I think what would be useful is for this committee to recommend that they actually do implement some of the areas that we want to actually get into. If we're going to get our technologies from the lab into the Fab, we do need government intervention, and that is where government is the only person who can act. So areas like that is where we see the.
10:12
10:13
10:13
10:13
Speaker 3
-
Speaker 3
So I think yes, it's about the support, partly money, but it's about the support. It's about the messaging. So as I say, I think things are a little better than they were three years ago because we now are actively discussing these things. It's about building international partnerships with the likes of TSMC, but also across Europe. So there are there are partners in Europe that we can work with. We've been working a lot with the Netherlands, with Germany, with Switzerland and others.
So we can build a regional strength. But international partnership is the heart.
So we can build a regional strength. But international partnership is the heart.
10:13
Speaker 6
-
Speaker 6
Of it. Final question from me. Right, which is that forgive me if this technology is so great, right? Holds out this incredible prospect to to turbocharge what we're doing against the the challenge that we're seeing, I mean, even reflected in the American elections about these vast data centers consuming entire reservoirs, forests, all the rest of it. Why isn't there some billionaire somewhere or trillionaire saying, do you know what? I can capture all of this value and effectively in the same way that TMS has become essentially a monopoly, I can capture that value too. Why? Why is the why wouldn't TSMC sorry say I'll.
How how fast can you get here? Right? How quickly can you get to Taiwan and show us what you got? So it's the resistance.
How how fast can you get here? Right? How quickly can you get to Taiwan and show us what you got? So it's the resistance.
10:14
Speaker 2
-
Speaker 2
There's no resistance. I did visit Taiwan in 2025 with the government and the Ukri, kind of like GB program. We did speak to TSMC and their version of innovate UK, kind of like catapult called Itri. They are working on the technology already, and if they do win, they win. That's it. That's it. Done. So it is something that we have the capability, we have the IP we need to get in now while they're still figuring out what they're doing. And we need to be fast. This opportunity is not going to be open forever. And once they have the PDK and they're.
10:15
10:15
Speaker 2
-
Speaker 2
So I'll give you an example. In 2025, I went I went with the UK advisor back at the time, I think he just retired for science and technology. He was in the room with the people from TSMC and I told them this exact ask. We have the IP in our universities. We want you to take this IP into foundries. They said, let us think about it. Right? They then asked me, who in the UK is going to take care of this? And at that point, we just figuring out what the UK Semiconductor Center was going to be.
So we didn't have an answer. I couldn't say as an individual, you're going to talk to me, TSMC, talk to me because I'm important. It needed that government body to be able to go there. Okay. The follow up from that, TSMC did put me in touch with the kind of German kind of like counterparts. We had a discussion, but it all came down to who are you to ask us to do all this? That that's what it comes down to. We are universities. We are not government. We don't have the power to be able to ask such big organizations to actually do us a favor, if that makes sense.
So it's up to you guys to actually to actually do it.
So we didn't have an answer. I couldn't say as an individual, you're going to talk to me, TSMC, talk to me because I'm important. It needed that government body to be able to go there. Okay. The follow up from that, TSMC did put me in touch with the kind of German kind of like counterparts. We had a discussion, but it all came down to who are you to ask us to do all this? That that's what it comes down to. We are universities. We are not government. We don't have the power to be able to ask such big organizations to actually do us a favor, if that makes sense.
So it's up to you guys to actually to actually do it.
10:16
10:16
10:16
Speaker 3
-
Speaker 3
Yeah. And I think they also want to see the end use. They want to be able to say there's a market there for it. Because if you think of TSMC, they make things for other people. So we want to get into that chain is yes, we want to speak to them, but we also want to have the people who are making things speaking to TSMC.
10:16
10:16
10:16
10:16
Speaker 6
-
Speaker 6
So I'm listening. I'm very excited about all this stuff. I mean, this is is radically new, having learnt electrical engineering in 1984, you know, and von Neumann machines were were exactly what we learned about. If I were Nvidia. I would see what you're doing and thinking you're about to eat my lunch and I would buy the lot of you. What's the protection against that. protection or why aren't they.
10:17
10:17
10:17
10:17
10:17
Speaker 2
-
Speaker 2
I'll give you an example actually of a company that I'm a big fan of. So fractal is a start up company that is based in the UK. It's run by Walter Goodwin. Yeah. When they started, they actually wanted to use some of the technologies that we're talking about today, but it found a level these technologies were not available. So then they did pivot. But if a company like fractal already knew that this is how they wanted to go or the line they wanted to go down on. So for them, they were trying to do computing memory technology, which basically is computing memory together, but not as neuromorphic, just close enough together.
And if they knew that they wanted to do this, and they've built such a company that has raised hundreds of millions of pounds, if we get that IP there for them and offer it to them, they could be the first people who could use this and actually make a company that is using most of the most novel memory technologies built here in the UK. So that's one example right there of like, why won't Nvidia get it? Because if we get our IP in there and then we back our own companies in the UK, then.
And if they knew that they wanted to do this, and they've built such a company that has raised hundreds of millions of pounds, if we get that IP there for them and offer it to them, they could be the first people who could use this and actually make a company that is using most of the most novel memory technologies built here in the UK. So that's one example right there of like, why won't Nvidia get it? Because if we get our IP in there and then we back our own companies in the UK, then.
10:18
10:18
10:18
10:18
10:18
10:19
10:19
10:19
10:19
10:19
10:19
Speaker 3
-
Speaker 3
Well spinnaker isn't isn't a spin out. Spinnaker is the system. And so that's an academic system. Spin cloud were then commercializing that. And they then took on the upgrade of spinnaker to spinnaker to scale it ten times. That's now in Dresden and Spin Cloud. I think they've just changed their name, but they're running that.
10:20
Speaker 1
-
Speaker 1
So. So yes, if you could let us know which companies have been targets and we can look at what the protections are in place. I think Kit's point is that is one of the kids points was why are companies like Nvidia not buying up all these great British companies? And I think we also need to look at whether the extent to which there is sort of a. Do you think there is a lobbying to prevent the investment. UK investment in these areas, given the likelihood of competition with existing very deeply entrenched business models? Do you do you think that's a possibility?
10:20
10:21
Speaker 1
-
Speaker 1
No, we recognize that. We recognize that, which is why we're asking about what protections are there and whether there are targets in the UK. I'm struck by the fact that that you you procurement, we're not you don't seem to be you're not identifying particular defense or other or departments which are should be procuring this. It doesn't seem like you have the ministerial or government support in terms of procurement. Would you agree with that? We're going to have to running out of time.
Would you agree that you don't have the.
Would you agree that you don't have the.
10:21
10:21
10:21
10:21
Speaker 1
-
Speaker 1
Okay. So you're not having the procurement pull through? No, not yet. And the funding doesn't have told us that it has was that there was 20 million of UK. There was £20 million funding. UK committed to neuromorphic computing up to 2030. Which doesn't seem. Is that will that deliver sovereign capability in this sovereign commercial capability in this key area?
10:22
10:22
10:22
Speaker 1
-
Speaker 1
So, so, so, so given. And then our report , A sovereign science diplomacy, we identified sovereign capability as being a key area, a key ambition for this government, given that the current funding will not deliver sovereign capability, is it is it fair to say that the UK's sovereign capability ambitions are not being funded in this area?
10:22
10:23
Speaker 1
-
Speaker 1
Great. So, mayor, did you want to make. Oh, great. Well, we've kept you a bit longer, but that's been absolutely fascinating. We've got some follow ups to. But thank you so much for joining us. And we have a brief break now of one minute for the next panel. Order. And welcome to this the second panel of the Science and Technology Select Committee's hearing on neuromorphic computing. And I'm going to ask Freddy to kick us off.
10:24
Speaker 5
-
Speaker 5
Thank you, chair. Thank you both for being with us. I'm a complete novice to this, so please talk to me like I'm an eight year old, which is always a good rule for politicians. Anyway, if you could briefly explain your organization, what you do, and if you could also start by helping to me, helping me understand what silicon photonics is, or photonic photonics, get it right doctor Josh Silverstone.
10:25
Speaker 9
-
Speaker 9
Thank you for having me. So I would like to give a little bit of my personal background as well because I think it feeds nicely into this discussion. So I'm a Canadian. I came to the UK to Bristol for my PhD about 15 years ago, and my PhD was really looking into how to build the largest scale photonic devices that have ever been built in photonic quantum computers. And so I joined the group of Professor Jeremy O'Brien in Bristol. I worked on the first architectures for photonic quantum computers ever with linear optics. Anyway, I then was part of the the kind of really early days of the creation of the IP that underpinned a company called quantum, which.
Is a bit of the one that got away, to be honest. But some of my colleagues really gathered up kind of all of the all of the foundational work that we had been doing in Bristol and ultimately took it to Palo Alto in California, because they couldn't see a way to raise the money to actually make that thing into a into a commercial reality that, that they recognized. So fast forward some years, I had a I had an academic career. I started my first company in 2016, control systems and really building up the underpinnings to allow us to build very, very, very complex photonic devices.
And so in our tag team today, I'll be able to provide the very biggest picture in architecture and how things fit together commercially, especially. So I started Hartley a few years ago. Hartley ultrafast again in, in, in Bristol is really is a neuromorphic computing company using photonics and really using that combination of analog, photonic electronic systems to deliver AI inference faster than anybody else on earth can. And so we are looking at applications really in spaces of the economy, where time is the absolute essence, whether that's in the financial system, whether that's across defense, whether that's in process control, in the network stack, you name it.
And so we are we've got a bunch of patents together pending pending in the US, and we are taking on a load of international investment now to really make these kind of dreams reality. It turns out that there are a great many similarities between trying to build a photonic quantum computer and photonic computer, you know, unadvised photonic computer. And we are really taking the kind of best practice, the tricks that we have learned, you know, collectively to making quantum computers work and applying those to accelerating AI in this way.
Is a bit of the one that got away, to be honest. But some of my colleagues really gathered up kind of all of the all of the foundational work that we had been doing in Bristol and ultimately took it to Palo Alto in California, because they couldn't see a way to raise the money to actually make that thing into a into a commercial reality that, that they recognized. So fast forward some years, I had a I had an academic career. I started my first company in 2016, control systems and really building up the underpinnings to allow us to build very, very, very complex photonic devices.
And so in our tag team today, I'll be able to provide the very biggest picture in architecture and how things fit together commercially, especially. So I started Hartley a few years ago. Hartley ultrafast again in, in, in Bristol is really is a neuromorphic computing company using photonics and really using that combination of analog, photonic electronic systems to deliver AI inference faster than anybody else on earth can. And so we are looking at applications really in spaces of the economy, where time is the absolute essence, whether that's in the financial system, whether that's across defense, whether that's in process control, in the network stack, you name it.
And so we are we've got a bunch of patents together pending pending in the US, and we are taking on a load of international investment now to really make these kind of dreams reality. It turns out that there are a great many similarities between trying to build a photonic quantum computer and photonic computer, you know, unadvised photonic computer. And we are really taking the kind of best practice, the tricks that we have learned, you know, collectively to making quantum computers work and applying those to accelerating AI in this way.
10:28
10:28
Speaker 9
-
Speaker 9
Yes. So photonics is the technology of light in the same way that electronics is the technology of electricity, right? Photonics means many things to many people. I think for the purposes of this discussion today we'll talk about photonics with respect to communications and computing. And in particular, what we're doing at Hartley is we are focused on that computing piece. So photonics for communications has been used since the dawn of time. You know, optical signaling if you like. But really that community or that computation use is a new one. And people haven't really cracked that yet.
10:29
10:29
Speaker 10
-
Speaker 10
Good morning everybody, and thanks for the opportunity to be here. So I'm Callum Littlejohns. I'm from the Cornerstone Photonics Innovation Centre. So we're the the UK's hub for the technology of silicon photonics. And our mission is very simple. It's to help innovators, researchers, entrepreneurs translate world leading UK research into deployed products, new companies and new jobs. So you asked about what is silicon photonics? So Josh kind of explained for tungsten technology of light and silicon photonics is integrating our little chip. So here I've got a little tiny chip that did come from a big wafer when it was made.
But what this can do for low energy computing is that it can transmit huge volumes of data, which much with much less energy than conventional electronics. So that's what makes it very compelling. And as we move to a world with physical AI , edge AI, that becomes very important. But the technology can also play a role in computing that Josh talks about, but also in sensing as well.
But what this can do for low energy computing is that it can transmit huge volumes of data, which much with much less energy than conventional electronics. So that's what makes it very compelling. And as we move to a world with physical AI , edge AI, that becomes very important. But the technology can also play a role in computing that Josh talks about, but also in sensing as well.
10:30
10:30
10:31
10:31
Speaker 10
-
Speaker 10
We should be extremely excited because, I mean, so there are projections coming from the US where data centers might consume 10% of their energy supply in, you know, 3 or 4 years relative to 2% today. That's because of the huge demand from AI. So if we can't solve that and make more energy efficient systems where photonics is the leading candidate to do so, then we need to build more power stations, find more energy.
10:31
Speaker 5
-
Speaker 5
So the first part of the panel we had was talking about neuromorphic computing. What's the difference between because they're both are saying that your your end goal is to reduce the amount of power needed and kind of energy that's used in advanced computing. So are you, are you tackling this from, from different angles? Essentially the same problem.
10:32
Speaker 10
-
Speaker 10
It's well, potentially the same. So for silicon photonics is the underpinning technology. So you might be able to build a neuromorphic system on technology of silicon photonics. So that's one application. We spoke about some others. So we're kind of tackling the same problem. Neuromorphic is one way of doing it. There are other ways as well. But so the area of photonics is the underpinning technology.
Josh can speak about computing and.
Josh can speak about computing and.
10:32
10:32
Speaker 10
-
Speaker 10
So it's being deployed now. It's Nvidia invested this year 6 billion across three major companies. Just two days ago there was an announcement Marvell, one of the world leaders, has procured, has done a deal to buy Lumentum, another one of the world leaders, and they value it in the announcement of something of the order of $115 billion. So they're buying Lumentum because they're one of the world leaders in integrated.
10:33
10:33
10:33
10:33
10:33
10:33
Speaker 9
-
Speaker 9
Sorry. Lumentum acquired a company called Aclara, a British company, about, I don't know, 6 or 7 years ago. And Aclara in turn. Was I mean, it was a rebrand, frankly, of a company called Bookham Technology that was the first company to ever commercialize Silicon Photonics and was a British company. So actually this this acquisition, I mean, it sounds interesting and makes headlines, but actually there is a UK angle to this to this story. Now, slightly alarmingly for me, you know, I, I have kind of eat and drank and breathed silicon photonics for, for a very long time.
And Aclara actually decided to wind down its silicon photonics activity after really buying up the world's first. And I think, I think we, I would love to address some of the, the kind of causality behind behind this stuff today.
And Aclara actually decided to wind down its silicon photonics activity after really buying up the world's first. And I think, I think we, I would love to address some of the, the kind of causality behind behind this stuff today.
10:34
Speaker 5
-
Speaker 5
I'll just ask one more question before before handing over why? Why are you in the UK? You're you know, you said you're Canadian. Bristol is a wonderful city. I studied there, I love it, but is that enough for you to be in the UK and to, you know, to make this a place, you know, where you build, you know, this technology and, and a company?
10:34
Speaker 9
-
Speaker 9
Yeah. I mean, it's why I mentioned it. I think, I think it, I think it is and I think it could be more so. But but yeah, certainly we have a lot going for us here. And I am relatively, you know, to, to latch on to some of the language around entrepreneurship recently. I'm relatively bloody minded about this. I am doing something like a, what we would call an a B test, you know, do one thing, compare it to another thing on whether you can build a big optical computing company ten years later from this quantum story that I told you here in the UK and ultimately companies like mine are only it's people, you know, we talk about whatever kind of regulatory environment and we talk about various resources and access and incentives and things like that.
But ultimately, you know, we are just people and the UK has a, a real, a real wealth of skill in this space. And furthermore different the other side of the Atlantic, you know, we potentially have access to a very large group of very bright people across Europe. And if I could, if I could make one point, it's that easier access to those people is, is a real benefit to
But ultimately, you know, we are just people and the UK has a, a real, a real wealth of skill in this space. And furthermore different the other side of the Atlantic, you know, we potentially have access to a very large group of very bright people across Europe. And if I could, if I could make one point, it's that easier access to those people is, is a real benefit to
10:35
10:36
10:36
10:36
10:36
Speaker 8
-
Speaker 8
Thank you. I'd like to talk about manufacturing and the capabilities in the UK, but maybe we ought to pick an application first. As far I'm afraid I'm not an expert either. But as far as I can tell, we've got quantum photonics. We've got, let's say classical non quantum photonics. And we've got integrated photonics that are a combination of classical integrated electronics with with photonic capabilities or functionality. Which of those is most mature.
Which is, which is, you know, most ready for deployment at this point?
Which is, which is, you know, most ready for deployment at this point?
10:36
Speaker 2
-
Speaker 2
Well, conventional integrated photonics for telecoms is being deployed. It's already in data centres around the world. But the underpinning technology for that is not too dissimilar to that you need for quantum as well. So and, you know, if there is infrastructure that you can build conventional silicon photonics circuits, it's almost exactly the same thing that you can build quantum circuits. And we see in and Josh mentioned Sai quantum earlier, they're using integrated photonics for their quantum computing. And UK based companies are similarly using integrated photonics. But the technology for quantum computing is still into the future.
But you know, it is being deployed elsewhere. So what we can do now in the UK, we're we're one of the world leaders in innovation research. The Photonics Leadership Group puts us third in photonics in the world behind the two usual suspects. Yeah. We can support companies through cornerstone doing proof of concept prototypes. For example, what I just showed you earlier. But evidence like university, we've spoken to more than 30 companies. We've done a survey of 100 companies consistently. We get the same message. The challenge is scaling up. We can prove something works. We can raise a few million, turning that into 100 million company is a massive challenge in this country.
We don't have the infrastructure to support it, and we can go to the world's largest foundries once we reach mass manufacturing stage. But there's a huge gap between one chip and one, you know, 1 million wafers. And there's a there's an infrastructure gap that needs to be filled there.
But you know, it is being deployed elsewhere. So what we can do now in the UK, we're we're one of the world leaders in innovation research. The Photonics Leadership Group puts us third in photonics in the world behind the two usual suspects. Yeah. We can support companies through cornerstone doing proof of concept prototypes. For example, what I just showed you earlier. But evidence like university, we've spoken to more than 30 companies. We've done a survey of 100 companies consistently. We get the same message. The challenge is scaling up. We can prove something works. We can raise a few million, turning that into 100 million company is a massive challenge in this country.
We don't have the infrastructure to support it, and we can go to the world's largest foundries once we reach mass manufacturing stage. But there's a huge gap between one chip and one, you know, 1 million wafers. And there's a there's an infrastructure gap that needs to be filled there.
10:38
10:38
Speaker 2
-
Speaker 2
Yeah. And that's already happening. So we can see. But these these massive mass manufacturing facilities would have standard processes. So they'll have something very locked down. Yeah. They care about yield and price. Yeah. So if your technology can adapt to that then you might be able to get in. I mean Josh can speak to that better than me, but if you want to try something different, you want to innovate. At the moment, you can only do that in a university and you can prove it works once. So what we need to be able to do is take that thing that we've proven works once, prove we can make it 100, you know, a thousand times, and then you've got a case that's compelling for the big foundries to start talking to you.
Without that intermediate stage, they're just not interested.
Without that intermediate stage, they're just not interested.
10:39
Speaker 8
-
Speaker 8
And in the previous panel, we talked about how, you know, the UK has got some niche capability in silicon. We're not at absolute zero. Where do you see kind of those seeds of potential in the UK at the moment, or are they not in the UK? You know, where are where are the seeds of possibility for us to develop kind of the silicon photonics?
10:39
10:39
10:39
Speaker 2
-
Speaker 2
Well, I run a facility where I use a facility at the University of Southampton as comparable facilities in Glasgow. So but then they're used by companies. There is a large amount of companies growing in the past few years, and we are seeing some significant investments in them. For example, a company called Salience Labs only, I think last month announced $125 million series B investment. I think they make optical switching, but they, you know, they use international facilities.
10:40
10:40
10:40
Speaker 8
-
Speaker 8
Yeah. I mean we're the question you know what we wanted to dig at is kind of I mean, we said in the first session, sovereignty is a word that means everything at the moment, but kind of that security and our resilience in being able to continue to progress this. But it sounds to me like that will come if we can scale. But scale is the big thing we need to start to pursue here.
Is that correct? How would you how would you describe it.
Is that correct? How would you how would you describe it.
10:40
Speaker 2
-
Speaker 2
Scaling up, not necessarily the mass manufacture. So that would likely still need international partnerships. But but who knows what the future holds. If you've got the scale up facilities and you've got all these companies like proving their products at huge volumes, you know, who knows what happens in the future. But something that's very important to make clear is that in integrated photonics, everything is much bigger than electronics. I mean, it's still relatively small, right?
10:41
10:41
Speaker 2
-
Speaker 2
The physical size of things. And that's very important because we spoke about ASML tool that costs £350 million to buy. Yeah. We don't need that for photonics. Okay. Everything is a te order, you know, at least an order of magnitude bigger. So we don't need the most advanced, extremely expensive equipment. That takes us 20 years to catch up. We need something with a lower resolution. So we do have the potential to do it in this country. In fact, we're doing it at very low volume, very flexibly already. So that's something that's quite important distinction to make.
We're not playing catch up in photonics in the same way we are.
We're not playing catch up in photonics in the same way we are.
10:42
Speaker 8
-
Speaker 8
Okay. That's really helpful. So when you look at the kind of really micro scale a couple of decades into Moore's Law stuff, where you it's incredibly specialist to work at that scale, you're saying our potential to build manufacturing capacity is actually very realistic. And despite the fact that we can see many, many orders of magnitude between where we are and where we could be, there are a few orders of magnitude we should be pursuing, and they should be within reach.
10:42
Speaker 2
-
Speaker 2
In photonics. Yeah. Because we're fundamentally limited by the way, the, the wavelength of the light we're using. We cannot make things smaller. They don't work. So we don't need to have the same drive for higher resolution as you do in electronics. And one thing to add is that there is already mass manufacture of integrated photonics happening in the UK, in Northern Ireland through Seagate. I think they provided evidence a year or so ago saying that they are actually Europe's largest manufacturer of integrated lasers because they use little integrated chips for their their recording heads for the hard drives.
But that's just for their products, right? That's not accessible to, to startups and other companies.
But that's just for their products, right? That's not accessible to, to startups and other companies.
10:43
10:43
10:43
10:43
Speaker 10
-
Speaker 10
Thank you. Yeah. GlobalFoundries is cutting a niche for itself really outside of this race to the very most advanced nodes that we've discussed with TSMC, for example. And so GlobalFoundries bottoms out at 22 nanometers of, of gate width or gate length rather. So our devices that we are , that we are developing today, we've sent some designs to them in, in New York state, they run out of Fab Eight in, in, in upstate New York. So Fab Eight has today a capacity of about 400,000 wafers per year. And they're looking at expansion to 1.2 million wafers per year.
And to put that in context, the kind of largest ambition that we can imagine in our computing space. And, you know, this is what we're talking about, is really high performance computing is in the kind of 200 wafers a month type 2000 wafers a year scale. And so even at the absolute kind of highest that we can imagine, our own volumes going out in product reaching, you know, it's an absolute drop in the ocean for a relatively globally, relatively small foundry, right? TSMC is going to be at least ten x up on that. So coming back to the, the question, which I spend now, much of my days thinking about is the value chain question.
And where does the value actually sit in this, in this, in this enterprise? Yeah. Yeah. Really. You can imagine that with capital costs, as we've discussed, are very expensive Dutch machines. And you can imagine with volumes in the hundreds of thousands or millions, the the margins on the foundries are very small. And so that actually leaves most of the value in the person who productize that, that system. And so we've taken a really conscious decision to be that product user and ultimately to capture the majority of that value in there, even though ultimately we're, you know, we're paying pretty big sums of money out to foundries to make batches of these semiconductors.
The ultimate value is in the product. And, you know, the product doesn't work without a design that allows that novelty to the end user. And the design has its value. Selling. That design, you know, by itself is, is one model that others pursue. But, but yeah, ultimately, you know, this isn't, it's not like some kind of magic new market, ultimately still, the value sits at the top of the value chain. And if you can, if you have the ambition to go and reach up there, then you can capture that and we can capture that here.
And to put that in context, the kind of largest ambition that we can imagine in our computing space. And, you know, this is what we're talking about, is really high performance computing is in the kind of 200 wafers a month type 2000 wafers a year scale. And so even at the absolute kind of highest that we can imagine, our own volumes going out in product reaching, you know, it's an absolute drop in the ocean for a relatively globally, relatively small foundry, right? TSMC is going to be at least ten x up on that. So coming back to the, the question, which I spend now, much of my days thinking about is the value chain question.
And where does the value actually sit in this, in this, in this enterprise? Yeah. Yeah. Really. You can imagine that with capital costs, as we've discussed, are very expensive Dutch machines. And you can imagine with volumes in the hundreds of thousands or millions, the the margins on the foundries are very small. And so that actually leaves most of the value in the person who productize that, that system. And so we've taken a really conscious decision to be that product user and ultimately to capture the majority of that value in there, even though ultimately we're, you know, we're paying pretty big sums of money out to foundries to make batches of these semiconductors.
The ultimate value is in the product. And, you know, the product doesn't work without a design that allows that novelty to the end user. And the design has its value. Selling. That design, you know, by itself is, is one model that others pursue. But, but yeah, ultimately, you know, this isn't, it's not like some kind of magic new market, ultimately still, the value sits at the top of the value chain. And if you can, if you have the ambition to go and reach up there, then you can capture that and we can capture that here.
10:46
Speaker 8
-
Speaker 8
So where do you think the bottleneck is then in this technology progression? You've we've gone into supply chains and you've said, okay, the UK has extremely small capacity and maybe we need to scale it, but actually you feel fairly confident that you can access manufacturing somewhere reliably. So what what's the bottleneck from your perspective in this deploying and scaling?
10:46
Speaker 10
-
Speaker 10
Yeah, that's I mean, so on the, on the kind of sovereignty resilience piece, I think that it, it's in, in photonics, it may be realistic to build a foundry. Certainly, I think it would be incredibly valuable for us to have a pilot line here in the UK. I think whether we have the capability to fully scale to our wildest dreams here in the UK, I don't see that future myself, actually. Okay. But.
But I'm sorry, there was a second part of your your.
But I'm sorry, there was a second part of your your.
10:47
10:47
Speaker 10
-
Speaker 10
Yeah. Right. So sorry. On, on, on the resilience piece, I think what we are focused on is diversifying our supply chain, right? So. We would love to have a supplier who can do something equivalent to what GlobalFoundries can do in New York State, perhaps outside of the US would be useful for us. You know, we're looking into defense applications, for example. And there is a lot of there is a lot of complexity around around that.
10:47
10:47
Speaker 2
-
Speaker 2
Can I maybe add. So over in the spring, we surveyed 100 companies in the UK who were looking to commercialize silicon photonics. So we've got some data that points to what the challenges are. Yeah. So I mean, the manufacturing of the wafers is one part. You also have to package it and integrate other systems. So 80% of the respondents said linking that supply together is a significant challenge. So off the back of that, we've proposed to bring all of that under one roof in at the what Josh called a pilot line. So that's like the scale up phase.
It helps companies to prove that it works in 100 wafers rather than 100,000. And it's also time to market at a bottleneck. So whilst you might be able to get into these facilities, you're always at risk of being deprioritized because it's international and inevitably they're going to be prioritising their own national companies. So 91% of the companies.
It helps companies to prove that it works in 100 wafers rather than 100,000. And it's also time to market at a bottleneck. So whilst you might be able to get into these facilities, you're always at risk of being deprioritized because it's international and inevitably they're going to be prioritising their own national companies. So 91% of the companies.
10:48
10:49
10:49
10:49
10:49
Speaker 2
-
Speaker 2
Know it's being deployed. People are in in data centers. Photonics has played a key role for a long time over long distances. We're now seeing the distances shrink. So that's what integrated photonics are all about , getting chip to chip interconnects and and the optics closer and closer. So so it's happening.
10:49
10:49
10:49
10:49
10:49
Speaker 1
-
Speaker 1
I think I remember photonics or optical connectivity coming to telecoms when I was working in the sector. And it has been quite extensively, obviously since then. But that does bring me just before I go to to Daniel. It's being deployed. It's an £18 billion sector. So do you have I mean, do you have orders? Do you have what is the government? Is the British government procuring from you? I mean, is that is the British government or the British private sector procuring from you? Is how how how is that driving the sector?
10:50
10:50
10:51
10:51
Speaker 10
-
Speaker 10
I if, if, if I require government to be there to buy my product, then I have failed. So government can be a bridge to markets which would be hard to access otherwise. But if government is the customer, then , you know, it's not that that that that doesn't work for me anyway. In fact, I think in the context of building really ambitious British businesses at this juncture where we have the end of Moore's Law, which is why we're all here worrying about this energy and speed and performance problem in compute. Moore's law has powered computing almost without anybody really thinking about it for 50 odd years.
And we are running out of nanometers. Yes, yes, yes to to, to build smaller and smaller digital transistors. And so that, that, that kind of physical motivation, it's not a, it's not a particular kind of market motivation. It's really physics that is leading us to now consider new ways of doing computing. And neuromorphic is something that has floated to the top there. Indeed in our.
And we are running out of nanometers. Yes, yes, yes to to, to build smaller and smaller digital transistors. And so that, that, that kind of physical motivation, it's not a, it's not a particular kind of market motivation. It's really physics that is leading us to now consider new ways of doing computing. And neuromorphic is something that has floated to the top there. Indeed in our.
10:52
10:52
10:52
Speaker 1
-
Speaker 1
Every 18 months. Yeah. Is is coming up to the limits of physics. And so that's what is another that's another reason why neuromorphic computing and, and photonics are huge opportunities. I'm just trying to get an idea of where, where, where your market is. You're saying it's not, you're not looking for government to be procuring silicon photonics or to support it.
10:52
10:52
10:52
Speaker 10
-
Speaker 10
But ultimately, yeah, we need to find customers in the in the big world. And I mentioned some of the verticals that we're looking at at the start there. Those verticals are most interesting to us because of the, the, the kind of advantage that we bring in time. You know, we are looking at really fast computing and but indeed, I mean, computing is, is technology that cuts across the whole economy.
10:53
10:53
10:53
10:53
Speaker 2
-
Speaker 2
Monopolized yeah, I can start and then wrong word. Sorry it is in the scale up phase. So we're we're great at proving things, you know, innovating, coming up with new things, but actually deploying that is a huge challenge. So it's, it's in that phase of proving that it can be manufactured reliably at big enough volumes that it becomes an interesting product. And that's what the data we've gathered is telling us that the companies are saying that that's what their challenge is. And I think it's not just us saying this. The Council for Science and Technology wrote to the Prime Minister, pointing to this as a as a challenge and recommending investment in a national photonics pilot line.
I think the AI hardware plan says similar things. Who identifies photonics as a UK strength? I think that's very clear. Identifies integrated photonics as strategically important to AI. I think that's also very clear. But that itself identifies pilot scale infrastructure as a constraint. And I'm quoting the the document there, but it also.
I think the AI hardware plan says similar things. Who identifies photonics as a UK strength? I think that's very clear. Identifies integrated photonics as strategically important to AI. I think that's also very clear. But that itself identifies pilot scale infrastructure as a constraint. And I'm quoting the the document there, but it also.
10:54
10:54
Speaker 10
-
Speaker 10
I think this is a complex system, right. You know, all of us together here. And I think that a pilot line fills a really important role in that system. And that role is certainly at the at at the earlier stages of technology development. It's we are able to deal directly with global foundries is incredibly unusual. Most companies of our size at our stage are not able to do that. And so they require engagement in other ways. Indeed, for some of the applications that we're looking at, it would be invaluable to have a UK manufacturing capability on this thing.
No photonics exists in a vacuum without electronics really. And so, you know, it's not not the final piece of the puzzle there. But, but, but no, I, yeah, if, if I, if I projected the, the impression that I didn't think it.
No photonics exists in a vacuum without electronics really. And so, you know, it's not not the final piece of the puzzle there. But, but, but no, I, yeah, if, if I, if I projected the, the impression that I didn't think it.
10:55
10:56
Speaker 2
-
Speaker 2
Is it's an infrastructure investment predominantly. So you need to build the facilities and it's capital intensive. So companies like Josh, you know, they won't be can't afford to build this on their own. So that's the role government can play in building a national shared infrastructure where it's managed centrally by a team. But the companies pay to access and they might buy 5% of the time or 10%. So you accumulate access to this facility.
10:56
10:56
10:56
10:56
Speaker 10
-
Speaker 10
I can I add add to that that thought, though, is that it's not just about taking risk, but it's taking risk here. If we want this to grow here, then we have to put our backs into it. I , I'm really confident that this capability exists elsewhere. I've engaged with it before, but it doesn't exist here.
10:57
Speaker 2
-
Speaker 2
And I think actually that's a very important point because it can anchor the companies here. I mean, we spoke about earlier what's stopping, you know, Nvidia just buying everybody. I mean, to be honest, there isn't much there's an nsai act. But if there's compelling reason to stay in the UK, of which infrastructure is one of those reasons, then they might invest in the companies, but they will stay in the UK and and the real value will be here in like the high value jobs and the R&D. And I'll give you an example of that.
It's kind of adjacent the spin out from the University of Southampton, making hollow core fibers that was bought by Microsoft. But they built a facility ten miles up the road because they want to access the fiber infrastructure that's local, and they want to access the skills and the talent. So that's the kind of anchoring effect that infrastructure can have. Not only do you have the infrastructure itself, it anchors the talent, it brings in investors. You know, you get a community of startups building around this infrastructure, and we see it in other countries like Imec and Belgium and prime example, huge scale up infrastructure covering all of semiconductors.
They got a massive start up environment built around them. We can do that here in the UK in photonics.
It's kind of adjacent the spin out from the University of Southampton, making hollow core fibers that was bought by Microsoft. But they built a facility ten miles up the road because they want to access the fiber infrastructure that's local, and they want to access the skills and the talent. So that's the kind of anchoring effect that infrastructure can have. Not only do you have the infrastructure itself, it anchors the talent, it brings in investors. You know, you get a community of startups building around this infrastructure, and we see it in other countries like Imec and Belgium and prime example, huge scale up infrastructure covering all of semiconductors.
They got a massive start up environment built around them. We can do that here in the UK in photonics.
10:58
10:58
10:58
Speaker 7
-
Speaker 7
Thank you so much. All sounds fascinating. I think from what I'm hearing from questions from my colleagues, we're struggling to see where you fit in. We all sort of roughly understand that today's computers, you've got Intel who build the CPUs, the chips. You've got IBM, as was who build the boxes. You've got Microsoft put the software on top. We understand those value chains. You're talking about specific applications. You're talking about building things that are fast. I'm picturing you as where Cray used to be, building supercomputers, actually finding those edge things, building them around so they don't have such big buses.
That doesn't take so long for the electrons to move around the machine building systems for share exchanges where microseconds matter. Is there a massive market there for I mean, your application specific devices is, is limited in the global reach. Are we looking at sort of medium scale companies that are coming out of what you're doing here being very specific, or is this a fundamentally technology changing issue across the entire value chain?
That doesn't take so long for the electrons to move around the machine building systems for share exchanges where microseconds matter. Is there a massive market there for I mean, your application specific devices is, is limited in the global reach. Are we looking at sort of medium scale companies that are coming out of what you're doing here being very specific, or is this a fundamentally technology changing issue across the entire value chain?
10:59
Speaker 10
-
Speaker 10
I, I mean, yeah, thank you for the opportunity to clarify this. I, I would say, and framed by the following discussion or the previous discussion, I think that we haven't had so much failure as much as a lack of ambition. And in fact, even luminosity, I would, I would put in that same bag. You know, those guys could have said, look, we're going to revolutionize the fiber infrastructure of the world, and we're going to replace every fiber in the ground with this amazing new stuff that we've got. But instead, they sold out to Microsoft, who put it in a bag.
And I don't know what they use it for now. And yeah, they probably got the Lamborghini in the nice house. But just the, the, the scale of ambition is different than, than the scale of ambition of Jen-hsun Huang went and started Nvidia and, you know, wanted to take over the world. And so coming back to your question about where does Hartley fit in this? Yeah, really. I want us to be a generational computing company that corners time as the thing that we can offer better than anybody else on earth. And, you know, as a physicist, all we have is time and space.
And we can, we can, we can be universally better in one of those two big cuts of the universe. So yeah, our applications that we can see in the kind of five year horizon are on the edge. So these are reducing latencies where it's critical could be on the markets. It could be in robotics. I think robotics is about to have a moment. There are big latency challenges in robotics today. Wherever the the time to get an answer to a question is the salient feature. We've been talking to fusion companies. You know, they have a reaction, a chemical reaction.
That is, it spins away faster than they can pull it back. And so really getting a faster bead on these kind of questions allows us to do more. We've been talking to satellite communications companies about revolutionizing the latency of their error correction. It means that they can reduce the power of their lasers or remove their satellites further apart, can cover the same earth with fewer satellites, this kind of thing, the biggest cost for them is launching heat sinks. So anyway, yeah, this this is where we fit. So absolutely many of our peers in the kind of future compute space, which is where we are now, I think is probably the hottest space in venture today , is certainly one of the hottest spaces of technology in the future.
Compute. Almost all of our peers are trying to be Nvidia one way or another, and we're doing something a bit different to that.
And I don't know what they use it for now. And yeah, they probably got the Lamborghini in the nice house. But just the, the, the scale of ambition is different than, than the scale of ambition of Jen-hsun Huang went and started Nvidia and, you know, wanted to take over the world. And so coming back to your question about where does Hartley fit in this? Yeah, really. I want us to be a generational computing company that corners time as the thing that we can offer better than anybody else on earth. And, you know, as a physicist, all we have is time and space.
And we can, we can, we can be universally better in one of those two big cuts of the universe. So yeah, our applications that we can see in the kind of five year horizon are on the edge. So these are reducing latencies where it's critical could be on the markets. It could be in robotics. I think robotics is about to have a moment. There are big latency challenges in robotics today. Wherever the the time to get an answer to a question is the salient feature. We've been talking to fusion companies. You know, they have a reaction, a chemical reaction.
That is, it spins away faster than they can pull it back. And so really getting a faster bead on these kind of questions allows us to do more. We've been talking to satellite communications companies about revolutionizing the latency of their error correction. It means that they can reduce the power of their lasers or remove their satellites further apart, can cover the same earth with fewer satellites, this kind of thing, the biggest cost for them is launching heat sinks. So anyway, yeah, this this is where we fit. So absolutely many of our peers in the kind of future compute space, which is where we are now, I think is probably the hottest space in venture today , is certainly one of the hottest spaces of technology in the future.
Compute. Almost all of our peers are trying to be Nvidia one way or another, and we're doing something a bit different to that.
11:02
Speaker 7
-
Speaker 7
We get very confused. I remember reading a report about how the Victorians were worried about the growth of horses in London, and London would be six foot deep in horse manure. So we get very. We think that today we'll carry on as it is. And you're describing a change in it, I think coming. I'm sort of struggling to find what we need to ask you. I think the biggest thing is what is the takeaway you want us to have to then form our ask to government?
11:02
Speaker 10
-
Speaker 10
Yeah. So for me, I think it, it, it all comes back to how we can increase the ambition of what we try and do. And I think that we have considerably more means than we give ourselves credit for. And I think that we need to be brave enough to, to go out and say, look, you know, we're going to do this across the whole earth and we're.
11:03
11:03
11:03
11:03
11:03
Speaker 2
-
Speaker 2
So I think the evidence we're seeing and that we've gathered is telling us that there's a scale up gap. So the ask is figure out how to solve that in a time bound like proper process to assess in silicon photonics, the pilot scale manufacturing needs and, you know, and decision milestones and decision criteria on that. Working with end users like Hartley and other companies who would use this facility to figure out exactly what they need.
11:04
Speaker 7
-
Speaker 7
We always used to look for not the magic bullet, but it's that that that product that leads it all. It's that I can't remember what we called it when I was working in telecoms, we'd look for the new, the new thing, the new text messaging, the new, the new killer app. And I think that's what we need to hear from to persuade government to invest in it, that this is the thing that will save defense. This is the thing that will save satellites. This is the thing that will save the space industry or whatever it may be.
And I'm not hearing that. Tom, did you want to come in on this?
And I'm not hearing that. Tom, did you want to come in on this?
11:04
Speaker 8
-
Speaker 8
Yeah. Thank you. I mean, I you mentioned defense earlier. I could see that being a very, very poignant area where we need fast decision making in complex problems. I just want to clarify that pilot line seems like a really good, tangible ask. If you were to define that line a bit tighter, your pilot line with a capacity of how many units per year and by what year?
11:05
Speaker 2
-
Speaker 2
Yeah. Okay. Good question. So it would be of the order of 10 to 20,000 wafers a year sort of scale. It would take probably 3 to 4 years at least to build something new. But we the what we've worked on with industry is building like a tangential R&D facility that could produce outputs much quicker. Yeah.
11:05
11:05
11:05
11:05
11:05
Speaker 6
-
Speaker 6
Yeah. I just wanted to pull together like a few things. Okay. Just very briefly, what is the future? Paint me a picture. We could have fusion, limitless power. We could have AI, limitless knowledge, and you could be producing limitless speed and capacity for computing. Quite exciting. And also slightly terrifying.
11:06
11:06
11:06
11:06
Speaker 1
-
Speaker 1
Great, great. Well, thank you very much. That's a very positive note on which to well, not quite end, but but limitless future. I just want to, to, to, to sort of part part of the ambition as well in our, in our report, as I mentioned earlier, we looked at science diplomacy in sovereignty. Now, I'm not sure I've heard it directly, but is this an opportunity for the UK to have sovereign capability in, in, in silicon in, you know, in, in, in a key part of the AI stack that we don't have now.
11:07
11:07
11:07
11:07
11:08
11:08
11:08
Speaker 1
-
Speaker 1
Okay. So so that's friendly manufacturers who share our values, as I think is the phrase or in countries that share our values rather. And you said, I think in the evidence I've seen said that you think this is going to have an impact in sort of 5 to 7 years. Is that is that what you were saying on the.
11:08
11:08
Speaker 1
-
Speaker 1
Doctor Silverstone nodding as I mentioned earlier, we have we have like trillions going into data center investment right now, and we have the IPOs of anthropic and and. Open AI coming, which is also involve trillions. You're talking about something which could fundamentally change the business model. Yeah. I suppose my question is so, so so that would create a lot of enemies, I imagine. My question is, is there, is there a way that that 5 to 7 years could become 2 to 3 years, or is that just physically impossible? And, and how are you, how are you and how can you compete with the, with the vested interests of the existing business model?
Just quickly, we've got a couple of minutes that quick question. You could try and answer that.
Just quickly, we've got a couple of minutes that quick question. You could try and answer that.
11:09
11:09
11:09
11:09
11:10
11:10
11:10
Speaker 2
-
Speaker 2
I think the impact of the pilot line might take that long because you need to build it, but the impact of the technology is happening now. And, and other companies are commercializing, deploying and selling this stuff. So, you know, if, if there were to be commitment for this, and that helps us to go and build these international partnerships for, for the mass manufacture, because they know what's coming. So that could help to accelerate.
11:10
Speaker 1
-
Speaker 1
Accelerate that talent. Well, I think that's I think that's really giving us something to, to think about if it's the right levels of investment we have, we have the capability to upend the existing business models for AI and, and, and compute. I think that's a, that's a really sort of positive story coming out and something that we should, we need to think about as a committee. We've taken over time a little bit, but that's been very interesting. And we've learned a lot, I think. And I just want to just rests for me to say thank you very much to our panel, our panelists, for sharing your expertise.
Order.
Order.