Steve Furber: Reverse Engineering the Human Brain (REPLAY)

Speaker A: Hello and welcome to the Danielle Noonan Podcast where I interview tech founders and innovators to learn the inspiring human stories behind the game changing tech we use every day. Today's guest is Steve Ferber and a seminal computer scientist, mathematician and hardware designer whose work includes the BBC Microcomputer M and the ARM UM32bit RISC microprocessor which can be found in over 100 billion devices today. Steve studied both maths followed by a PhD in Aerodynamics at Cambridge before joining Herman Hauser and Chris Curry at Akon Computers. For the next decade he would work with a first class team of engineers and designers to to revolutionize the home computer market before he and Sophie Wilson went on to design the ARM processor with a relatively small team and um, budget and with little inkling of the consequence it might bring to the world. In 1990, Steve left Akon and moved to Manchester where he is now professor of Computer Engineering. At the university. He was charged with leading research into asynchronous systems, low power electronics and neural engineering which led to the Spinnaker project, a supercomputer incorporating a million ARM um, processors which are optimized for computational neuroscience. He's basically trying to reverse engineer the brain. A lofty ambition even by his own admission. In this wide ranging conversation we discuss Steve's life journey from studying maths with professors such as the famed John Conway and Sir James Lighthill to the highs and lows of building the BBC micro and and the story behind the ARM 32 bit RISC microprocessor. I thoroughly enjoyed talking to Steve and I'm overly excited about his Spinnaker project which we also discuss. So here is my conversation with Steve Ferber. Steve, thank you so much for joining me today. I always like to start with people's backgrounds. So I wanted to ask you, what were you like growing up? What were your hobbies and interests?

Speaker B: So my interests growing up were fairly narrow. I was brought up in a house in Marple. My father was a mechanical engineer working for the nuclear power group, um, on the early British nuclear power stations. My interests in my teens were model aircraft. I wasn't very good at it, but I kept persisting. And some basic electronics I guess. I built, um, a little board that enabled me to wire valves together to form interesting circuits for various functions. And I had electronics kits. I also inherited a fairly large, uh, Meccano set from my father which I still have.

Speaker A: Wonderful. So your father was an engineer and I know your mother ended up teaching maths. That was something you were passionate about what did you do your degree and your PhD in?

Speaker B: My undergraduate degree was in maths at Cambridge. I lent towards the applied side, uh, of maths. And then, um, for my Ph.D. i moved to the engineering department and my Ph.D. was in aerodynamics, although towards the mathematical side of aerodynamics. So it wasn't a complete, um, change from my undergraduate degree. It was just a shift of emphasis.

Speaker A: I always like to reach out to people who know my guests, know them better than me, and ask them if there's something I should ask. And I reached out to Tudor Brown and he said to ask you about active sound cancellation, which he said was something you studied. Is that right?

Speaker B: To some extent, yes. In fact, it wasn't my PhD topic, but, um, the PhD student with whom I shared a room and a supervisor worked on some very early anti sound work. Um, he, uh, built a system to try and cancel noise emissions from things like electricity substations. So this was quite large scale and this is achieved by basically measuring the sound output and trying to generate the inverse field to cancel it out. And he was quite successful at this. And of course that's the basis now of things like noise cancelling headphones.

Speaker A: That's interesting that Tudor got it slightly wrong, but I mean, we're talking about so many years ago. That was the other thing. I wanted to speak to Sophie. And then I was thinking, you know, it's a very long time ago that you were all working together at acorn, but can you tell me what that time was like? Because you had Sophie, who I mentioned Sophie Wilson, but you also had Chris Curry and Herman Hauser. What was that motley crew like? How did you all meet each other?

Speaker B: Okay, this is a long story, but in the course of my PhD and my subsequent research fellowship, which was also in the area of aerodynamics, I joined a university society called the Cambridge University Processor Group, which was a student society for students who liked building computers for fun. And, uh, in this society, you know, the real men built their computers out of TTL, which was a standard logic family in the 1970s. And, um, the WIMPs like me use these newfangled microprocessors. And I built computers, um, initially for fun, and then I started using them in my aerodynamics research for data logging and so on. And ultimately I wrote my PhD on the computer where I'd, uh, built the hardware myself and developed the text editor that I used for writing the PhD. So, um, I'd become an enthusiastic amateur in the computing field. And when Herman and Chris got together and decided they wanted to start a consultancy company in the microprocessor area. Um, then they knew of the University Processor Group and they thought this was an obvious place to go and find people to help them with their, uh, um, computing consultancy work. So I was approached by Herman in my room on Trumpington street, across the road from the engineering department and, and he said they were forming a company and was I interested. And I, I expressed some interest in having an involvement, although there were limits, uh, as to how far I could go because of my research studentship and my research fellowship and the terms and conditions of those.

Speaker A: So that's obviously why, because I know you were doing this fellowship which meant that you kind of work part time with Herman and Chris, but do you think this Processor Group was obviously very instrumental? Was it like the homebrew compute in the States where Steve Wozniak, Steve Jobs, they all came together there? I'm guessing it was a similar type of club.

Speaker B: It was very similar, yes. And I was not one of the founders of the Processor Group, but I went to the first meeting when they called it. My interest was, was slightly motivated by, um, my interest in flying. I mentioned I worked with model aircraft in my teens and that actually continued for the next decade or so. Um, but I'd begun to realise, having spent a, ah, year in the Cambridge SPIDER Club, that flying real aircraft took an unreasonable amount of time. So I started thinking about building flight simulators and the obvious basis for doing that would be small, uh, computers. So my interest in the Processor Group was linked to my interest in flying and aerodynamics. That's how I got, uh, drawn into that. In the Processor Group, we did very scary things. We ordered microchips from California on credit cards which were very newfangled and students. We didn't have a lot of money so this was high risk and we hand built machines. A lot of us used a technology called Vero wiring, which you had a standard circuit board which you could put chip sockets into and then you could wire them together using a kind of wiring pen, um, which used insulated wire where the insulation melted when you soldered it. So you could basically knit your chips together on the back of the board and assemble machines that way. And that's the way I built my early computers.

Speaker A: That's fantastic. And also said you're a bit of an amateur, but it doesn't sound like that. I mean, what, what year are we talking about?

Speaker B: Uh, I guess we're talking about 1978-79. My memory on dates is not particularly clear, but I think I must have submitted my thesis in 79. The PhD was awarded in 1980. So about that time.

Speaker A: Okay, and you were there in the early days of Acorn. When did the BBC Micro Project come about?

Speaker B: The BBC Micro project started in, I'm pretty sure it was early 1981 when Chris Curry picked up on the rumors that the BBC were looking for a computer to use as the basis of their series of TV programs called the Computer Program. And I'd been involved at Acorn in developing a concept for a next generation machine to follow on from the success of the Acorn Atomic, which we called the Proton Machine. And um, when we were invited by the BBC to show them a prototype, we did some quick adjustments to the Proton and built a prototype in a week. It's been fairly famously documented. And that was the basis for the BBC Micro. And I, I think that was, we put that together around Easter 81. This was still before I was an employee of Acorn. I was still doing things out of interest with them in exchange for being given components for building the Mach I needed for my research in the engineering department.

Speaker A: When I've watched. Is it called Micro Men, Is that right? Yes, yes, yes, I love that. And then I actually watched you, Herman and Chris talking about it afterwards and it was all rather amusing. It was obviously an exciting time. And the BBC Micro project came rather fortuitously because it ended up giving you obviously a complete, um, timeline of what you were going to be doing and what you needed to do before the opportunity arose. What was the plan for Acorn? I know you mentioned, mentioned the atom. What was the Proton going to be?

Speaker B: Acorn had started from the consultancy which Chris and Herman called Cambridge Processor Unit, or CPU Limited. And um, within CPU Limited, we actually did a little bit of prototyping for Science of Cambridge of their MK14 product. And um, Sophie Wilson looked at this and said, I can do better than that and went home and produced the design which she called the Hawk, but which became the ACORN system 1. So ACORN had started by selling very basic hexadecimal keyboard, hexadecimal display. It was actually not a single board computer, it was two boards, but effectively at the single board computer level. And then those got expanded into racks that were used for various sorts of professional control applications. But the racks, the racking and the supplies and so on made them quite expensive. So then Chris Curry organized the development of the Atom, which was effectively putting a lot of this technology in a much cheaper format inside a box with a, with a Full keyboard. And the Atom was Acorn's, uh, first single box computer product. The Proton was intended to take this further. And we realized at that point that new microprocessors were coming into the market. Uh, the atom used a 6502 which was an 8 bit processor. And um, 16 bit processors were coming along, but it wasn't quite clear which direction Acorn should take with those. So the Proton was originally designed as a dual processing system. It used a 6502 front end, um, to use all the code that we developed for the Atom to handle things like the keyboard and floppy disks and so on. But then it had an interface we called the tube to a second processor which we anticipated would be a more powerful 16 bit microprocessor of some sort. So that was the concept that was in the air. Now when the BBC arrived, it was clear that their target price point couldn't be met with the full dual processor. So the BBC micro evolved from the front end 6502 of the proton, but it retained the ability to connect second processors. Later, uh, people familiar with the BBC Micro will know that it had a tube interface and you could connect a range of second processors through that tube interface.

Speaker A: And what was the competition like? Because obviously this was in a time even before people called them personal computers, their home computers. Who else was doing similar work? Or who was your competition? I'm assuming obviously Sinclair.

Speaker B: Yes, I think we had a rather unhealthy focus on Sinclair as competition and vice versa. I think that's probably correct. But I mean, I think that focus was partly because of course Chris Curry had come from Clive Sinclair's territory. He'd worked for Clive before forming Acorn. But, but there were, there were actually a very large number of home computer companies around in the UK at the time. I'm not sure I could remember all of them. A lot of them seem to be named after fruits, so Tangerine and was there an apricot? I can't remember. Uh, there were of course the home computers coming from the States, from the Tandy and the Apple, uh, Atari. Um, was that a computer or just a game? I'm not sure. Commodore. So there were actually dozens of companies, uh, building small, uh, simple home computers at the time. And um, I think in terms of the wider market, all of these were new companies that people had not heard of and were not familiar with. And what was achieved with the BBC Micro was attaching a very well known and respected brand, that of the BBC, to a home computer, which meant that people who were not so sure which of these newfangled companies they might be able to trust, they saw the BBC logo and assumed that was trustworthy. And that contributed a lot to the market success of the BBC Micro, I'm sure.

Speaker A: And I know that originally everyone was thinking it was going to sell. I think the number was 12,000. I don't know how that number came to be, but obviously it sold way above more over a million. How did you get the contract? What was it that you guys had over St. Clair?

Speaker B: I think obviously you get a more accurate response to that if you ask the BBC why they chose Acorn. Um, but I think, um, in that week, you know, when, when Herman did his, uh, phone calls saying, can we build a prototype in a week? And I told him not to be ridiculous, and then he sort of told little fibs and basically said, sophie thinks we can do it. So I had to agree to have a go. And, and, and, and we built it in that week. I think the BBC had been working with a different company and they got quite frustrated with the rate of progress. And when they saw that Acorn could do in a week what other companies took months not to achieve, um, that was one thing. But in the same week that we built the electronic prototype, if you like, Herman, um, also got Alan Boothroyd to build a case prototype. So when the BBC came, they could see on the one side what the machine would look like and on the other side what it would work like. And I think they were reasonably convinced by the fact that Acorn could move quickly and get things done. And of course, Acorn had already got the Atom on the market, so we were not a complete newcomer to the game. Um, so I think it was that balance of forces. Um, there was also the issue as to how flexible the company they worked with was going to be about meeting various fairly complex aspects of the specification that the BBC was aiming for. And I think they got the sense that Acorn was going to be very flexible with them.

Speaker A: Um, what would you say were some of those difficult aspects? Because I know that what BBC wanted and what was achievable might have not met in the middle to start with. But what were the more harder technical aspects that you guys took on and were able to obviously achieve?

Speaker B: I think one of, one of the more challenging aspects was the teletext receiver. The BBC Micro was developed, of course, before the age of the Internet. Although the Internet roots can be traced back to the 1960s. In the beginning of the 1980s, the Internet was certainly not widely available. And so the issue of distributing software was quite high on the agenda. And the BBC had this idea of distributing software through redundant spaces in the Teletext transmissions. Building a suitable teletext receiver so that you could plug a TV signal in and extract programs that were broadcast by the BBC through Teletext was one challenging aspect. Another aspect was that they wanted lots of I O ports, lots of places where you could plug things in. And the BBC Micro had a lot of interfaces built into it. Ultimately their spec was to, uh, look for a machine that, based on a Z80 running CPM, which of course the BBC Micro itself couldn't do, but that could be delivered through a Z 80 second processor and was delivered. Um, so there were all these aspects, but also they had very strong views about the language. They wanted the machine to run the language which became BBC Basic. They didn't want a standard, relatively simple implementation of Basic of the sort that was in widespread use at the time on other home computers. They wanted something that was a much richer programming environment. And it was clear that they thought that ACORN could deliver that.

Speaker A: Why was that? Why didn't they opt for the simpler option?

Speaker B: I'm not sure. Again, you'd have to ask them that. But I think it was partly because they wanted to use the machine as an educational platform. Um, so they wanted the language running on the machine to have the sort of desirable properties that a reasonable computer language should have. Uh, so they didn't, for example, want it to, uh, be heavily based on things like go to statements which are if you use them excessively, you end up with programs that are very hard to debug. Um, they wanted a version of BASIC that had better structures.

Speaker A: Okay. And with the timescale that you had, I know it was done fairly quickly. What were some highlights for you? You obviously had this fantastic team. You're fairly young, you're fairly green, even though obviously you'd worked on a computer before. What were some highlights for you around being in that team at that time?

Speaker B: I think the highlight for me, you know, I'd been in academic research for a few years and I was finding the sort of openness, uh, of academic research rather difficult to work out. You know, what it was you should do day to day. Whereas when I joined ACORN staff, the commercial priorities were extremely clear. So it was for me much easier to feel that each day was being spent productively than I found it previously in academic research. But obviously there were good people involved in ACORN at all levels. We were very well Supported by Chris and Herman as the managers of the company. Um, they gave us a lot of freedom to make decisions ourselves and a lot of support in implementing those decisions. And it was a very small and tight team at the time. Acorn grew very quickly on the success of the BBC Micro, but I think when we started the project, the company was no more than about 20 people

Speaker A: and it grew phenomenally quickly. What was the impact when it grew? And it obviously became a huge success, the BBC Micro, and it went into many, many schools here in the uk. What was next? What did you want to do next? And how did you feel the company changed as the company grew?

Speaker B: Yes, the company did grow very fast. And, um, um, for the first few years we had very well defined objectives to deliver on, um, by way of second processors. And there was an attempt to extend the BBC Micro into the professional arena through the ACORN Cambridge workstation, which was a BBC Micro with a second processor attached for doing more serious computation. So obviously a company changes its nature very significantly, um, when it expands from a few tens of people to several hundred, um, and, uh, the internal management structures have to become more formal and procedures have to be set up, so on. But I think all that was Acorn took in its stride and did pretty effectively. Um, of course, it did run into financial difficulties with an attempt to move the BBC Micro technology down market with the Electron. Um, that ended up being rather expensive. And also the company tried to establish international markets, including in North America, which turned out to be very expensive as well. So commercial reality started to bite a few years in, and that gave the company a very different feel, I'm sure.

Speaker A: And the other thing is, I'm guessing this was Herman and Chris's first company. So there's one thing being innovative and getting a whole bunch of engineers together and working on something, but then being able to grow it and see new opportunity. Like you said, the commercial side of it, it can prove extremely difficult. And obviously it was at a time when I think you started the company, when people were very interested in home computers. Then there was obviously a lot of competition coming in. Uh, mentioned North America. I know that was difficult time for the company trying to make it over there. When did you decide to work on a new chip? When was the moment when you were moving away from what you had been working on to focusing on that?

Speaker B: This was 1983. So the BBC Micro was launched beginning of 82 or the end of 81, depending on which account you read. And we were busy through 82 developing the things we were committed to the sort of second processors and so on. In 83 we began to settle back a bit and say, well, the BBC Micro's done very well for us. Um, how do we build on this into the future? What's going to happen next? And at that point it was even clearer that 8 bit micros were reaching the limits of their capabilities, that the market would be moving to 16 bit microprocessors. So during 83 we spent quite a lot of time looking at the 16 bit microprocessors that were on the market and um, actually deciding we didn't really like them very much. They had technical drawbacks that meant they were not doing as good a job as we thought they should. And so we, we began toying with the idea of what would a processor look like that didn't have those drawbacks. And I think Herman presented us with papers from California on reduced instruction set computing, where we saw that a graduate class at UC Berkeley had developed, um, one of the first risk processors that was highly competitive and they'd done it with very limited resource. And at that point we began to think, well, maybe risk is the way to go and maybe we can uh, consider developing our own microprocessor. Now at that time it was really a paper exercise and we expected that if we started out on this course that mainstream industry would pick up on the risk idea and trample all over us before we got to the end. So we were treating it initially as we more of a learning exercise than actually a product development. Um, but we carried on with the exercise and mainstream industry didn't pick up on risk. And we found ourselves with a prototype arm chip on April 26, 1985 that plugged in and worked rather nicely. Now by that time the project had become more formal. We had plans for a machine that had four custom chips, of which the ARM was just one. Acorn had started a research lab in California led by Jim Mitchell, developing a new operating system called arcs. And by the time we had the chip, we were on a much more formal footing within the company. So that's the period of time from sort of towards the end of 83 when we started seriously developing the arm, to April 85 when we had the first chips.

Speaker A: I think it's incredible because I did read the process was about 15 months. What surprised you most? Because obviously it was something that happened, I'm guessing m, uh, pretty simply and quickly compared to what you thought might happen. What was, was the most surprising out of all of that?

Speaker B: Well, I think the most Surprising thing is that mainstream industry did not pick up on this risk idea until much later in the 80s. And so the ARM was the first functioning RISC microprocessor and risk delivered on all its promise. So we had our understanding of what made a processor effective and having high bandwidth access to memory was the key. And that's why ARM, um, uh, started as 32 bit. So ACORN effectively went from 8 bit to 32, skipping the intermediate step of 16 bit processing. And we were really quite surprised when it all came out and the chip worked straight off and delivered straight off. Um, we were even more surprised when we started measuring its power consumption and discovered that it was, uh, extremely low indeed. So that the processor was very power efficient. We'd deliberately designed it to use less than a watt of power because that's the limit for low cost plastic packaging. If you go much above a watt, you have to go to much more expensive ceramic packaging. So for cost reasons we aimed below a watt. Um, but actually when we measured those first chips, they were coming in under a tenth of a watt, which for sort of state of the art performance was very surprising.

Speaker A: But also this is not something that you and Sophie had worked on before. What gave you the confidence that it could work?

Speaker B: I don't know, some crude form of technical arrogance. I mean, one of the things that helped of course, was that, uh, UC Berkeley published quite a lot about their risk processor design and what their graduate class had produced. So we had some examples. I mean, ARM is not a Risk one copy. It's quite different from RISC one, uh, because we had somewhat different priorities. Although it's very much inspired by the principles underpinning RISC 1. But we saw how that class had put things together. Uh, we were reasonably experienced logic designers. We built semi custom chips on for anti ulas for the BBC Micro and on a bigger scale for the electron. We still, when we started, thought there was some mystique to designing microprocessors. But as we just went through the process, what we found was that there is no mystique. A microprocessor is just a piece of logic like all the other chips we designed to go on the previous machines.

Speaker A: And when you look back at that time, what are you most proud of?

Speaker B: Well, I'm personally proud of the fact that we sowed the seeds for what has become, I think, the world's most successful microprocessor. Of course that success has been built through subsequent generations and the company arm itself and thousands of people have worked to build that success. But you, uh, know, ARM Limited would not be there without the seeds that we sowed in the 1980s. So I'm pleased with the way that's turned out. But actually also there are things uh, about what the BBC Micro achieved in its time that I also find rewarding. I keep having people come up to me and say it was the BBC Micro that converted them into programmers and they built their careers on programming from what they learned on the BBC Micro. And I think the BBC Micro did have quite a significant impact on a generation of programmers and engineers that been quite significant in terms of the economy. So I'm um, pleased with how the BBC Micro impacted the UK in particular, but also some other countries. But obviously I'm also very pleased with how ARM has turned into a global success in the subsequent decades.

Speaker A: Yes, indeed. And the thing about the BBC Micro, I mean, saying it was quite a significant impact is a huge understatement. And there are, like you said, a generation of people that grew up and got their chops on this BBC Micro. And I've heard even people like Demis Hassabis, he said that he learned on a Spectrum zx. But there is that culture around young, well, boys and girls obviously who grew up with a computer in their household. And every engineer I know has stories and every kind of innovator I speak to talks about how they were able to have a computer accessible not just in school but at home. And it's changed the trajectory of how the tech industries even move. So it is quite the accomplishment to be involved in both. I wanted to ask you because obviously arm, um, spun out into a separate company. Was Apple involved in setting up the company or how did it work?

Speaker B: Well, before ARM was set up as a company, I left Acorn just before I moved to the chair at the University of Manchester that I still occupy. I'm uh, currently in a semi retired capacity, so I wasn't involved directly in the story. But for my last couple of years at Acorn with Malcolm Byrd, who's the technical director, we looked at ways to spin out the ARM activity because it was clear that Acorn's market was not growing fast enough to support the kind of development that ARM needed to stay competitive. So it looked at ways to try and spin uh, off the ARM development activity, but none of them had worked financially, if you like. So what I understand is very soon after I left Acorn was approached by Apple, um, who had been developing this product called the Newton, which was a handheld PDA product and they'd been working with at And t planning to use the ATT Hobbit processor, but they'd become frustrated with where the Hobbit was going. And they already knew about the ARM because the ARM had been used, uh, through a license from PLSI technology in graphics accelerators for the Apple II and so on. So they had some familiarity with the ARM and they decided they wanted to use it in the Newton, that they were not comfortable with using it while it was owned and controlled by a competitor, uh, albeit a competitor that looked rather small from Apple's boardroom. So they came to Acorn proposing that the ARM Activity be spun out into a joint venture. And of course, they were knocking on an open door because this is what Acorn had been trying to achieve. Uh, they want, you know, Acorn was keen to offload the development cost of the arm, they were keen to keep ARM development going, but they were not keen to pay for it. So Apple offered them a solution that was mutually convenient. Uh, VLSI technology, who made the early ARM chips, also came in. Um, and, um, ARM was set up as a result very quickly. I mean, ARM was established in the barn in Swaffham Bulbeck by November that year, when I had left before any of this had started at the end of July.

Speaker A: If this had happened, the deal with Apple had happened several months earlier, would you have stayed on there?

Speaker B: That's a very good question. I don't know the answer to that question. I often say that, uh, my decision to return to academia may be the most expensive decision I ever made. On the other hand, if I'd stayed and joined the ARM Activity in the company, maybe I would have ruined everything, so maybe it wouldn't have happened. Who knows?

Speaker A: Well, yeah, this is the thing. No one knows. And also, I think the reality was you had a young family at the time and, you know, we, we make decisions based on what we need to do at the time. But obviously it wasn't like you went away and then did nothing else. I mean, it's amazing what you've gone on to do. I want to ask you, because obviously you went to Manchester, like you said. Can you give me a very brief history of computing in Manchester? Because I actually didn't know when I was doing my research for you. I'd heard of the baby, but I didn't know it was in Manchester. And I just wanted to know if you were able to give, like, just a minute overview of kind of the history of computing in terms of Manchester.

Speaker B: Uh, describing Manchester's history of computing in a minute is a Challenge because it's a very deep history. But yes. Freddie Williams and Tom Kilburn built the first programmable computer, the Manchester Baby, that ran its first program on June 21, 1948. And that was the first machine to implement Alan Turing's idea, uh, from the 1930s of the Universal computing machine. Shortly after Baby was operational, Turing came to Manchester and started using the machine. And of course while he was in Manchester he wrote his seminal paper on, um, computability and intelligence, where he proposed the Turing Test for human like AI. Now following the baby, Manchester carried on building a series of big machines, um, a transistor machine in the 50s. And then probably the most notable Manchester machine was the Atlas in the early 60s which was the world's first supercomputer. It was responsible for several inventions, most significant of which is virtual memory. Um, so the idea that the programs and data you can have on your computer are not limited by the size of the memory. That was a Manchester invention for the Atlas. But Atlas also, um, had hardware, floating point and many other, uh, innovations. And it was the fastest computer in the world when it first ran. Manchester then carried on building machines, I guess, uh, another Highlight was the MU5 machine which was the prototype for the ICL19, uh, hundred series of mainframe computers which underpinned the UK local authorities and central tax and so on for, for a couple of decades. And after that the, the nature of the research at Manchester changed a bit. But uh, before I arrived, Ian Watson and John Gerd had led the development of dataflow machines at Manchester. Manchester with mit, was in the vanguard of the dataflow revolution. I went in 1990 and through the 90s our focus turned to designing chips upon which machines could be built. And we built asynchronous versions of the ARM processor through the 90s which were branded the Amulet series of microprocessors. And then from the turn of the century my focus has been on the Spinnaker project, which is using massive numbers of ARM processors to uh, implement brain modeling computers. And the Spinnaker machine incorporates just over a million ARM processes and can support real time models of significant brain regions or potentially one whole mouse brain.

Speaker A: I honestly found it so fascinating when I was looking into this. I wanted to ask you before we get into how it has come to be on your work in the last 20, 30 years, but what can you tell me about why you wanted to do this? What was your mission behind it?

Speaker B: So in the 90s, I, uh, returned to academia in 1990, at the beginning of the decade and, um, had some connections with research funding, particularly from the, the eu. I took some funding from ACORN with me to Manchester and I'd been reading about asynchronous processes and was fascinated by the idea of building, uh, processors and computer systems that didn't require a central clock to cause them to operate. So we worked on that through the 90s and showed that you could build pretty much anything you wanted to in an asynchronous style. The Amulet series went from Amulet one to two to three to three synthesized sparkors. Um, and there are a number of advantages which accrue from running asynchronously. Um, but there was the disadvantage that asynchronous design was not well supported by industry standard design tools. So it became increasingly clear that it was going to be a hard sell to persuade industry of the merits of this asynchronous design approach when they couldn't find a source of efficient tools to work that way. So that's why my direction switched a bit around 2000. Um, I'd also by that point been involved in the design of computers for over 20 years and they'd become maybe a thousand times more powerful than when I started. But they still struggled to do things that humans, even young humans, find quite straightforward. So I became increasingly interested by the differences between computers and brains. There are similarities. They're both information processing systems, they both receive inputs and generate outputs, but how they turn those inputs into outputs is quite different. And so since 2000, I focused on using, um, computer engineering skills to see if we can build machines to give us more insights into how the brain achieves its formidable capabilities.

Speaker A: What's so interesting about this is you've gone onto this deep mission into solving one of our oldest problems and trying to work out how the brain works. When we look at intelligence in terms of human beings, what is it? What makes humans intelligent? Because how do you reverse engineer the human brain without us understanding it?

Speaker B: That is a good question. But I mean, it's like most scientific quests. You, you start with partial understanding. You use that partial understanding to try and build experiments that yield further understanding and then you can fold that further understanding into the next experiment. So, you know, you never, you never start in the scientific quest by knowing the answer. And, and the brain is like that. We do know a great deal about, um, what the brain is composed of, the neurons, uh, glial cells, synapses that connect neurons. We know quite a lot about the low level details and we know quite a lot about the different regions of the brain. And we have hypotheses about the different functions of those different regions. So it's not that we know nothing at this stage. Um, but what we don't know is one of the fundamental principles of the operation of the brain as an information processor. How is information represented in the brain? How is it distributed across all these neurons? How is it stored? How do we, you know, recognize a face? As one of the challenges which only fairly recently, computers have become quite good at, um, and m, it's not clear that computers do it in a similar way to the brain at all. So there are plenty of big questions to go at. Setting off with the goal of understanding how the brain works is still over ambitious. Um, I often quote Ada Lovelace, who wrote about her, uh, interest in generating a theory for how the brain works. Uh, she was thinking about this 200 years ago. Um, but even today that's over ambitious. Um, but what we can do is take a brain region, see if we can model it, see if the model will reproduce biologically measured data, and then see if, having built a model that seems to capture at least some of the aspects of the biology, can we then use that model to understand what's happening? That's the way this research progresses. And for the last 10 years, we've been doing this work in the context of the European Union Human Brain Project, which is a massive project involving over 100 universities across Europe, with neuroscientists at one end, uh, theoretical neuroscientists, computational neuroscientists, neuromorphic computer people such as ourselves, but right through to medicine, um, what do we understand about diseases of the brain? Can we use IT systems to aggregate data that give us new insights into how to manage and perhaps even cure those diseases? So it's a huge interdisciplinary activity. And, uh, that's been a very exciting context in which to do this work for the last 10 years.

Speaker A: It's incredible. Where do you see it going in the next 10 years? If you were to predict, which I know would be impossible, but if you were to fast forward 10 years, what would the spinnaker be doing? What would you have learned, do you think, or hope?

Speaker B: I do hope that the spinnaker machine will, in collaboration with neuroscientists and others, um, will lead to new insights into how the brain does its job. But in parallel with the 20 years of development of Spinnaker, there's of course been this explosion in industrial AI, which is based on, um, neural networks, but not the same sort that we model on spinnaker artificial neural networks have exploded into use and we're now using them all the time in our everyday lives and their capabilities are very impressive, but they also have limitations and there's now a growing expectation that more brain like computing might overcome some of the limitations of conventional neural networks in engineering applications. So I think for the next decade there's going to be a big focus on trying to work out where brain inspired, uh, computer chips, uh, can contribute to engineering applications and commercial activity in that space. And Spinnaker is poised to contribute to that. We have a second generation chip that we've co developed with the Technical University of Dresden and they have set up a company in Dresden to commercialize Spinnaker 2. There are many other companies engaged in similar activities. It's not yet clear where the winners will be, but I do think there's going to be a significant increase in focus on commercial applications of brain inspired computer chips alongside steady progress in using those brain inspired computers to add to our knowledge of the brain itself.

Speaker A: Like I said at the beginning of this, it's such a fascinating area. What would you see, you said there about the commercial aspect? What's the commercial side of it?

Speaker B: Do you mean where does this technology work commercially?

Speaker A: Yeah, because it's not being used. Is it commercially now? Is that right?

Speaker B: It's being used in very limited ways. Where I feel it's likely to find commercial application is in what we call very energy sensitive applications at the edge of computing, in the context of Internet of things, where we have distributed sensors throughout a building or throughout a bridge or whatever, doing AI processing close to those sensors to reduce the amount of data that has to be sent to a central computing resource to minimize the energy requirements and bandwidth requirements. I'm expecting to see brain inspired computing find a role there. It may find a role in other areas where current AI solutions are being used, but their energy costs are, are a problem. So there may be a role, for example in self driving cars or in driver assisted cars where there's already quite a lot of neural network technology being used for processing images from cameras to either control the car or to assist the driver in controlling the car. And those applications could potentially be done with much lower energy requirements using um, more brain inspired neural networks.

Speaker A: Thank you for explaining this and thank you for your time. I've got one question which basically I start the interview at the beginning and I like to circle back there at the end. I wanted to ask you what's one piece of advice you'd give yourself?

Speaker B: I think the, the main Advice I offer to people who are at an early stage of their, uh, careers is unless you have a very clear idea of what you want to do, you know, if you want to be a doctor, you know, go and study medicine and go and be a doctor is fine. But most people at the early stage don't have a clear idea of where they want their, uh, career to go. And so my advice to people in that position is to make decisions that keep as many doors open as possible. So study a subject at university that has pretty broad application and, and uh, I studied maths and that's quite a good basis for a whole range of careers in science, technology, engineering and possibly even medicine. I never regret my decision to study maths as an undergraduate because I think it provides a foundation for a very wide range. But it's not the only subject. People tend to only study maths if they feel they've been fairly good at it at school. School, um, I think studying most branches of science, certainly physics and most branches of engineering, studying computer science equally is highly flexible. So as I say, just choose subjects that keep doors open rather than narrowing your future opportunities unless you know exactly what you want to do.

Speaker A: No one knows exactly though, do they? The amount of times I remember as a child my grandparents every Sunday would come round and asked me and my siblings what we wanted to do. And my brother and sister seemed very intent on m. I think my brother wanted to be a lawyer, my sister wanted to be, I don't know, physiotherapist or something. And I had no clue and I always felt inadequate for not knowing. We're talking about like, you know, being 12, 13, 14. And of course my brother never became a lawyer, never studied law, my sister never went into physiotherapy. So it's one of those things, in hindsight that you're absolutely right. You've got to keep your options open. Unless you have a clear path, path in front of you. Keeping the path wide and open is a good idea. Just one last question actually, which is not a question I normally ask, but as someone who has followed this path that you have when you look back at the BBC Micro, which was obviously part of the BBC's initiative, the computer Literacy project. What would you do differently in schools now to get children excited about engineering and computing?

Speaker B: I have been involved in a couple of Royal Society studies of computing in schools, um, which have done in depth analyses of how computing is currently taught and we've made recommendations which have actually led to significant curriculum changes 10, 12 years ago a lot of school computing was focused on ict, which was basically using office productivity tools, which is a useful skill, but it's not terribly exciting. And we've tried to get more of the core of computer science into the school curriculum, which has much more substance and interest to it. And that's been partly successful. I mean, I think the kind of things that are happening today that feed into this very effectively are. An example is the Raspberry PI movement, um, where they really enabled access to very impressive computer technology at very low cost. So anybody who's interested in can probably afford a Raspberry PI and it's a complete computer where the basic ones start at about 20 or 30 pounds and you can do things with that and schools can form clubs. So I think there are a number of people looking at this, uh, quite carefully who are doing the right things and how you persuade more people to be interested is a very difficult question. And the school curriculum is a very complex subject. In particular, of course, in the context of computing, how we persuade more girls to be interested. Because there's still a very significant gender imbalance in UK undergraduate intake for computer science and there's no reason why there should be. There's nothing about computing that suggests it should be a male dominated subject.

Speaker A: I agree. I think actually on that note, I think it's representation or a lot of it is because I worked in startup startups from 2006 and I remember thinking, God, there aren't many women at these conferences. And you know, I started researching and I, I ended up writing a book called Female Innovators at Work because I just felt at the time, and this was in, uh, the mid-2000s, that the press weren't writing about female engineers or women that had run tech companies. They just weren't writing about any women, full stop. And so it became one of these bugbears of mine and ever since I've tried to interview as many as I can. And actually when we're talking about Newton earlier I interviewed Donna or August who led the engineering team at, ah, Apple. I think there's a lot of women who have done significant things in technology and they're just not written about. So, you know, that's still, still my thing to focus on. But anyway, all right, thank you so much for your time, Steve. I really appreciate it.

Speaker B: You're welcome.

Speaker A: Thanks so much for listening to my conversation with Professor Steve Ferber and thank you to Steve for graciously giving me his time. As I said at the top of the interview, I am overly excited about the Spinnaker project, which I will link to in the show, notes the idea that they have built a massively parallel chip multi processor system for modeling large systems of spiking neurons in real time is quite incredible. As Steve said, the machine incorporates a million arm processes linked together by a communication system that can achieve the very high levels of connectivity observed in biological neural systems. Such a machine is capable of modelling up to a billion neurons in real time, which despite still only being around 1% of the human brain, is still amazing. But before you go, as always, I want to leave you with a quote. So today's comes from Alan Turing in his Computing Machinery and intelligence paper in 1950, which we mention in the interview today. Turing says we may hope that machines will eventually compete with men in all purely intellectual fields. But which are the best ones to start with? Even this is a difficult decision. Many people think that a very abstract activity like the playing of chess would be best. It can also be maintained that it is best to provide the machine with the best sense organs that money can buy and then teach it to understand and speak English. This process could follow the normal teaching of a child. Things would be pointed out and named again. I do not know what the right answer is, but I think both approaches should be tried. We can only see a short distance ahead, but we can see plenty there that needs to be done.