Fungible Compute, Mortal Computing and Why the IoT won't use Silicon

Thin film transistors, smart packaging, and edge intelligence with Richard Price of Pragmatic Semiconductor

Hey all! A god to honest interview for you today. If we can all stop thinking of AI for a god damn minute. Let’s go to the other end of the spectrum. Not trillion dollar token factories in the sky. But cheap, ubiquitous computers in every object.

As we all know, the semi industry spends its energy pushing performance upward. Smaller nodes, faster transistors, more compute per watt. But as discussed, this means the cost of entry keeps going up. A cutting-edge fab costs $20–40 billion. Only three companies in the world can manufacture at the leading edge. This has been the defining dynamic of semiconductors for decades: fewer players, higher stakes, more concentrated capability.

But what about the other direction? Not faster chips for data centres, but cheaper chips for everything else. The vast majority of physical objects in the world — packaging, labels, agricultural products, wearable patches — have zero computational capability. The Internet of Things was supposed to change this. It largely hasn’t, because silicon chips are too expensive and too rigid for disposable, flexible, or ultra-low-cost applications.

Pragmatic Semiconductor, based in Durham in the UK, is building an alternative. Instead of silicon, they use indium gallium zinc oxide (IGZO), a material that’s been used in display technology for decades, deposited as thin films on flexible polymer substrates. What you get is a chip that bends, costs a fraction of a silicon equivalent, and can be manufactured in a fab that fits in 20 by 30 metres. Process times are measured in days, not months. The facility already produces billions of chips per year, with room to scale five times within its current footprint.

In previous State of the Future interviews, we’ve explored the computing stack from multiple angles — Synthara’s compute-in-memory to eliminate data movement at the chip level, Phanofi’s coherent optics to make data movement efficient when it’s unavoidable, Wave Photonics’ GaN PICs. Pragmatic represents a different vector: pushing computation outward to objects that have never had it.

What Did I Learn?

1. It’s useful to think about a bifurcation. The bleeding edge will keep pushing to 2nm and beyond, but the bigger untapped market might be the trillions of physical objects with zero computational capability. Pragmatic’s IGZO-on-polymer approach isn’t competing with TSMC. It’s a new-ish category. Maybe the IoT won’t be silicon?

2. Manufacturing speed changes the economics of everything downstream. Process times in days mean lower inventory, smaller fabs, and the ability to deploy manufacturing at customer sites. Pragmatic’s 20-by-30-metre modular fab is as much a strategic asset as the chip design itself.

3. Edge intelligence doesn’t need to be sophisticated. It needs to be cheap and everywhere. Tiny classifiers running on a few hundred gates won’t replace cloud AI, but they’ll capture data and make simple (increasingly sophisticated) decisions The value here is the aggregate data layer not the sale of an individual chip. Fits neatly with Dan’s Mortal Computing thesis. But goes one layer deeper in that, if chips get cheaper, you push even further into the concept of “fungible compute”.

---

The Interview

Lawrence: Richard, give us the quick version. What is Pragmatic, and what are thin film transistors?

Richard: I’m co-founder and CTO at Pragmatic Semiconductor, we founded the company 15 years ago. Thin film transistors are essentially field effect transistors that, instead of using a bulk semiconductor like silicon, use intentionally deposited thin films of semiconductor materials. In our case, it’s n-type metal oxide semiconductors.

Lawrence: And this spun out of Manchester, right? 2010?

Richard: The origins are actually a little earlier. It originally came out of some research at the University of Manchester, looking at novel types of semiconductor device designs and alternative thin film materials. That’s where I met my co-founder, Scott White. That business didn’t quite succeed in the first instance. But Scott and I saw the nucleus of some ideas around smart packaging, the ability to use thin film semiconductors on objects, exploiting the form factors. We had some early commercial interest, so we actually set the business up in Cambridge. It’s often mistaken as a spin out, but it’s actually more correctly a spin in. We took the opportunity to Cambridge and started working with the university there.

Lawrence: Why Cambridge specifically?

Richard: It’s a combination of things. The talent in Cambridge, a network of both research and businesses that have been working in similar areas, a lot around display technology, which has some similarities. There was a logic to move into Cambridge at that time, primarily from working with the university and seeing that we could build talent there.

Lawrence: So break it down for me. When most people think of a chip, they think silicon. What’s different about what you’re doing?

Richard: The difference is you can deposit the thin films very quickly and cost effectively. That’s really the foundation of a lot of display technologies — the backplanes in displays use similar sorts of processes and materials, but at much larger scale. You can create arrays, circuits, and build up the foundations of semiconductor devices: transistors, switches, capacitors, resistors. And then what we’ve built on top of that is the more classic interconnects that you’d be used to within a silicon chip — the back end of line wiring that allows you to put those devices together and create circuits.

Lawrence: And the specific material is IGZO — indium gallium zinc oxide. It’s been in displays for decades. What makes it interesting for circuits?

Richard: It’s got a higher electron mobility than materials like amorphous silicon, which were historically used in displays. But it’s also got a very low off state, very low leakage. And that, actually, for things like DRAM, is attracting a lot of interest — looking at hybrid integration with things like CMOS and adding this capability on the back end.

Lawrence: Help me understand the flexibility part. Is the bend coming from the material itself or from the substrate you’re putting it on?

Richard: It’s a combination. The enabler is the mechanical support, the substrate, which is a polymer — we use a polyimide — and the thin films that you construct on top of that aren’t thick and brittle, so they’re able to flex and bend in conjunction with the substrate. There’s research going back 20-plus years on concepts of foldable mobile phones. A lot of that in the early days was around polymer semiconductors, and one of the challenges was getting the performance and lifetime to match product requirements. Then newer classes of materials came through which had higher performance but were still able to maintain flex and bendability.

Lawrence: So when I think about flex ICs, I shouldn’t be thinking about putting them in data centres competing with GPUs. We’re making a new class of semiconductor for things that don’t currently compute. Is that the right framing?

Richard: Yeah, it’s essentially starting to merge the physical and digital worlds. We’re not looking at competing with bleeding edge semiconductor nodes going into data centres. We optimise the functionality for what’s required for the product. In our first generation of products, these are NFC-enabled chips. You can read them with smartphones or other NFC readers. These allow you to globally tag or provide a unique code to any object. You put those on consumer goods — household products, bottles of water, food, beverages, perfumes. And that allows you to interact with consumers, do anti-counterfeiting, brand promotion, loyalty campaigns. A whole range of things unlocked by that unique code embedded onto a physical object.

Lawrence: Most people know they can tap their phone for payment, maybe they’ve got an AirTag. How would the average consumer understand what you’re making?

Richard: They’d be more familiar with a contactless payment, Apple Pay, Google Pay, which uses the NFC interface in smartphones. This allows them to use that same interface to interact with products, redirect them onto the web, a unique URL specific to an individual item. Over time, we’re adding sensing capabilities — temperature, humidity, chemical sensing information. We’re building up increasing sophistication of functionality. Things like data logging. You can do this on a pallet level now, but being able to do data logging of temperature on an individual item could be very valuable. Take something like a vaccine. On the package level, it makes sense to track the temperature, but if you can do that individually, you can make sure you’re not wasting viable products and you’re able to have information specific to an individual item.

Lawrence: I can imagine the supply chain use cases. But the sensing part is what gets me. I’d love to have temperature and humidity sensors in every room of my house, but it’s too expensive. Does flexible IC help solve that cost problem?

Richard: Certainly that’s part of the unlock. In some of these cases, there’s an elasticity between price and volume. If you can reduce that price, the volumes increase massively. We see areas like smart agriculture as well, being able to get information maybe at the plant level, where you can then optimise irrigation and when you might add nutrients, to even more efficiently grow and optimise yields. The combination of form factor and the ability to manufacture at really high volume — we’re already manufacturing in the UK for these kinds of products in the billions of units, with the ability to scale to at least five times that capacity just in our existing facility in Durham.

Lawrence: How? How are you making these so cheap? Talk me through the manufacturing.

Richard: First, any semiconductor manufacturing is not straightforward. It requires very reliable, proven manufacturing equipment. We have tools in our fab in the UK that you would see in any fab in the world, including TSMC. They’re well proven and designed to run 24/7 with high reliability. What we do with that is we use different materials, and each of our process steps is very short. Because we’re using thin films, the time to do a process step is very short. Actually, the bulk of our manufacturing cycle time is queue time — it’s wafers waiting for tools to become available to go on to the next step. So we can actually manufacture with a raw process time of a few days to make a chip.

Lawrence: A few days. What does that look like in steady state?

Richard: It’s longer, but we’re talking weeks rather than months. And that also allows us to reduce the footprint of our fabs because we don’t have as much work in progress. Our fab is essentially modular in design, it’s 20 by 30 metres, and from that we can do billions of chips. It’s a really compact design, and that means it’s more energy efficient, and uses obviously less carbon as a consequence.

Lawrence: 20 by 30 metres. That’s the size of a tennis court and a half. And you’re producing billions of chips from it. OK. So why build it in the UK? I hear constantly that the UK has high energy costs, it’s not a manufacturing hub. Why Durham?

Richard: A few reasons. The footprint of our fabs is relatively small, so actually it’s not as energy intensive as pretty much any semiconductor fab. Yes, we would like to see lower energy costs, they’re a contributor. But they’re not as punitive as they are for some people. From another perspective, we’re British as a business, and we’ve been here for 15 years, and we want to develop the core of the technology and our manufacturing base here. Part of it is a desire to make this work in the UK, and that makes some things a little harder. But that’s definitely our intention.

We’ve been able to attract the talent that we need through a range of routes, including repatriating people that worked in the semiconductor industry in manufacturing in the 1980s and early 90s, some of whom were already in the region, recruiting internationally, and developing a talent pipeline. I think it is possible, and we want to make it work. The government recognises energy costs are too high. I’d like to see quicker movement on ways to bring those down as a broader benefit to the UK economy. But it’s an important part of the mix for us, not the critical decision maker at this point.

Lawrence: What about the cluster argument? Saxony gets thrown around a lot. Are you fighting a good fight alone up in Durham, or is there a supply chain building around you?

Richard: I would actually take the UK as a cluster. I think we’re small enough not to be thinking about regional clusters. If you look at OEMs and chemical suppliers, they’re going to think UK-wide. There are other manufacturers in the UK that use some of the same suppliers, same chemicals. The majority of people we work with, there are European hubs — some in the UK, but many in mainland Europe for the OEMs. We have local support that’s usually only an hour or so away. I don’t think the UK is in that bad a position.

If you look more broadly — there’s the compound semiconductor cluster in South Wales, Seagate in Northern Ireland that’s been around for a long time and is an often untold success story. There’s still lots of activity in Scotland. Photonics in areas like Southampton, design strengths around Bristol and Cambridge, and a pretty strong academic community distributed around the country. We don’t have large fabs in the UK. But we’ve got significant strengths in quite a number of niches and an opportunity to build on those.

Lawrence: You currently operate as a hybrid IDM — you design and manufacture. Is that the long-term model, or does this evolve?

Richard: We’re manufacturing and designing our own products now. We see that trend moving more to foundry over time. But there’s a hybrid opportunity because we have this very compact manufacturing footprint. We also see the opportunity to deploy our manufacturing at customer sites. We would operate the fab on behalf of customers, but they would design the products. It’s a bit of a hybrid model.

One of the reasons we’ve had this specialisation in silicon is in large part because the cost of the research and development, and the capital cost of deploying new fabs as you’ve gone to more advanced nodes, has just increased astronomically. You go from tens of players being able to do manufacturing down to only three that are able to do it. It becomes a challenge because you need 20 to 40 billion dollars to deploy a fab.

Lawrence: The idea of deploying a fab at a customer site — that’s a genuinely different model. You can’t ship a TSMC facility somewhere. But 20 by 30 metres, that’s portable. Let me ask about applications beyond smart labels. You mentioned wearables and AR/VR?

Richard: We’re working with customers around miniaturisation, using our flex IC essentially as a smart substrate. You can do fine line interconnects and then build on top of that systems — integration of silicon electronics and surface mount components. Over time, take some of those capabilities and integrate them into the substrate itself. Things like resistors and capacitors can reduce the number of surface mount devices, reduce the BOM, and make the whole system smaller, more flexible, and lower cost.

There’s quite a lot of market pull in wearable devices, things like AR/VR headsets where volume actually becomes a driver — not just footprint, but the total physical volume that the electronics occupies. If you can shrink that down not just in x and y, but also in the z axis, and make it more flexible, you’re then able to deploy that with a better form factor in devices that require flexibility.

Lawrence: What about healthcare? That feels like a natural fit for something flexible and cheap.

Richard: We’re actually seeing real opportunities in healthcare. You’ll have seen things like continuous glucose monitors emerge in the last several years, increasingly moving to a consumer product. The ability to make something even thinner and more comfortable, at a cost point that allows it to be democratised — available not just in the developed world, but also in economies that don’t have sophisticated healthcare systems.

Things like brain-computer interfaces and other types of healthcare wearables where the combination of the flexibility and something that doesn’t have rigidity allows you to get a better interaction with the body, to conform and move with the body when it’s being worn. I think we’ll see more in that direction. And it’s something I’m really passionate about from a personal perspective. We’ve been working on a number of proof of concepts for several years.

Lawrence: There’s a lot of focus right now on sovereign AI, strategic semiconductor independence. Is there a story for flexible ICs in that narrative, or is that trying to put a flat peg in a round hole?

Richard: It depends a little on definitions of AI. Essentially, what we’re allowing is capture of additional data from a whole range of different environments and objects. That data will feed AI. We have the ability over time to do very simple decision-making or machine learning at the edge or the item, and to enable some of those decisions to be pushed to the edge. So you’ve got less requirement to move data up the stack. We see opportunities there. But as we talked about earlier, we’re not doing cutting-edge GPUs.

Lawrence: Right. We’re talking about relatively simple classifiers, not distilled LLMs running on your devices.

Richard: Certainly not LLMs as they’re currently imagined. We generally look at optimising the circuit design, optimised for the specific job in hand. That allows you to strip back functionality that you don’t need. We actually published something last year on tiny classifiers where we’re using an evolution algorithm to optimise the circuit design. In some cases, you can reduce the complexity of that down to a few hundred gates for the task in hand.

Lawrence: A few hundred gates. That’s beautifully minimal. Are there any objects you’ve put your circuits inside that might surprise people?

Richard: A lot of the early demonstrators were things like beer bottles, so probably not surprising. But we’re seeing real opportunities in healthcare, as I mentioned — CGMs, brain-computer interfaces, other wearables. The combination of the flexibility and a semiconductor device that’s inherently flexible opens up a whole category of applications where the electronics can conform to the body rather than sitting rigidly on it. I think that direction has a lot of potential.

Debrief

This interview series has, without quite planning it, been mapping different layers of the computing stack. Synthara and SEMRON are rethinking computation at the memory level, stopping data movement before it starts. Phanofi is making the movement that remains as efficient as possible with coherent optics. Wave Photonics is working on the photonic integrated circuits that could redefine how light carries information on-chip. All of these operate at or near the data centre.

Pragmatic is working at the other end entirely. Not faster computation for centralised AI, but dispersed, purpose-built computation for the physical world. The connective thread is the same question: where in the stack can you add intelligence, and what do the economics have to look like for it to make sense? At the data centre, the answer involves billions of dollars in capital expenditure on cutting-edge fabs. At the item level, it involves fractions of a penny on a chip manufactured in days.

The healthcare angle is what stuck with me most. Continuous glucose monitors that are thinner, cheaper, and comfortable enough to wear without thinking about them, available in countries that can’t afford the current generation. Brain-computer interfaces where the electronics flex and conform to the body. This is where flexible semiconductors can have real impact beyond packaging.

The deployable fab model is the other idea I keep coming back to. In semiconductor manufacturing, scale has always meant centralisation: bigger fabs, more capital, fewer locations. Pragmatic’s compact footprint inverts that logic. Ship a fab to a customer site, operate it on their behalf, and you’re not just selling chips, you’re offering manufacturing as a service, distributed rather than centralised.

One question lingers. How big can the edge intelligence story actually get? Tiny classifiers on a few hundred gates are elegant, but the gap between that and useful autonomous decision-making is big. The near-term value is clear, smart labels, sensing, unique identification. The long-term vision of purpose-built intelligence on every object depends on use cases that aren’t just technically feasible but commercially justified. Pragmatic has the manufacturing story figured out. The next chapter is proving the world actually wants billions of intelligent objects, not just billions of smart labels.

For now, the Durham fab is humming. Billions of flexible chips, manufactured in days, going onto objects that never had computation before. If the future of AI depends on richer, more diverse real-world data, someone needs to build the capture layer. Pragmatic is making a credible case that they’re it.

---

Check out Pragmatic website for more, and Richard is here.

Bye.