 Felly, dyna'r dda wedi bod wedi'i wneud o'r cyfnod i'r ffordd. Yna yw'r prinsfyniad hulwt i'r prinsfyniad. Mae'r ddysgu'r cyfnod i'r ddechrau. Mae'r ddaf yn eilio 42 yn roi i rhynodd. Mae gennym i'r ddalasig i'r ddechrau o'r cyfnod i'r ddysgu sydd, yn y system yw rhaid i'r ddysgu, sy'n cymrydio'r cyfnod i'r ddysgu, a ddyn nhw'n ddysgu'r ddysgu'r ddysgu. ac mae'r sgwr o'r sgwr yw wedi'i ddweud â'r amser o'r pethau. Rhywodd â'r llyfr i'r pethau. Felly yna'n gweithio'r mae'n rhai amser o'r pethau oedd ymddangodd, wrth gwrs? Rhaid i'n rhaid i'n rhaid i'n rhaid i'n rhaid i'r pethau. Yn 1992 o 3 o'r rhaid i'r pethau o'r pethau o'r pethau o'r pethau o'r rhaid. Mae'r arm ddechrau yn wneud nhw yn 1983 ac yn ddechrau yn y rhan ni barnwyr 1985, dwi gyrdestru'r yr arm yn ei ddweud yn ei gweithio'r ddechrau'r argynnerau ac mae'r arm yn ymddangos yn ei ddweud, felly mae'r arm y ddweud yn seithrwyr ar eich computereb IBC, a rydyn ni wedi'u gweld ddweud arneud agorwyr yn 1986, fel ddefnyddio'r unig, ac oedd ydy'n ddwyig yma i erbyn yw 1987. Cyn oedd y gallu ei wneud o'r dda i'r ddweud yma. Felly mae'n defnyddio? Felly, dyna'r ddweud y Fydd Wilson. Felly, dyna'r ddechrau o'r ddechrau'r ddechrau'r ddweud. Felly, gallwn yma'r ddweud? Mae'n ffordd o'r ddweud? Felly, mae'n ddweud bod hynny'n ddweud ar hwnnw i'r ddweud. Felly, after the BBC computer, ACORN's first really major success, not our first actual success, and the ACORN Atom and the ACORN system range were strong successors enough for us to be able to start a company and keep going, we were looking around for a processor to replace the 6502 which was in the BBC computer and there wasn't what we wanted. In the BBC ond we built a four mega transfer per second memory system and we gave half of that to the processor and half to the video system. So we're quite capable of designing memory systems with high performance and we sort of wanted a processor to use that performance and couldn't find one. Felly, dyda dwi gweithio'r rhych. Felly, dyda dwi gweithio'r rhych. Mae'r rhych yn ei bod yn ei bod yn erioedd, i fod yn ni elu'r rhychiau? Dumwyr, dyna dwi gweithio'r rhych yn ei bod yn ei bod yn ni. Ond yn ymreithio am 400mh652, unol 8286, 06800, gallwn llwanido nôr 3216. Mae'r ddafydd o'r ddafyn am hyn o'r cyfnod am hyn oedd eich cyfnod, ac yn fwy o'r ddafyn ar gyfer ychydig, mae'n gweithio'r cyrraedd gweithio yn fwy o'r ddafyn. Nid yw i'n gwybod i'r cyfnod oherwydd, oherwydd, dyma'r eu cyfle i chi'n gweithio'r ddau, sydd oedd eich cyfnod a'r cyfnod o'r ddafyn yn gweithio'r ddau, ond yna'n gweithio'r ddau cyfnod o'r ddau o'r cyfnod o'r ddau. Er y gallwn fod yn ddatud o ddwyicio'r ddwyciadeth o 4 MB per sefyllurent, oeddo dawnodd mynd i yn frankin cynhyrchu lleol yw'r ddwyciad ddwyciad hwnnw o ddwyciad hwnnw o sut mae'r ddwciad ddwyciad hefyd yn 4 MB per sefyllurent Dyna'r 4 MHz 6502 o ddwyciad hwnnw o ddwyciad o 4 MB per sefyllurent ac unwys yn 8 MHz ddwyciadau 3616 i weld y 4 cychwyn bod lefnod rhagaf ac dyddo ddwyciad arall ddwyciadau ar y llif, per sefydl, ond mae'r cyfnodol yn 16 bit. Felly mae'n cael ei wneud o'r 4 megabytes. A byddwn ni'n ddweud o'r prosesau sy'n cymdeithasol ymlaen nhw. Mae'r cyfnodol yn ychwanegau, boi'r cyfnodol sy'n cymdeithasol yn ffast o'r cyfnodol a'r cyfnodol sy'n cyfnodol yn ffast o'r cyfnodol ac mae'n cyfnodol yn ffast o'r cyfnodol yn ymddiol yn ffast o'r cyfnodol. A byddwn ni'n ddweud o'r cyfnodol sy'n cymdeithasol eisiau yn ffast o'r cyfnodol i'r cyfnodol yn ffast o'r cyfnodol ac mae'n rhai phoedd ymlaen. Mae hi ddim yn yma. Felly mae'n cael ei ddweud o'r cyfnodol sy'n cyfnodol i ddweud o'r cyfnodol i ddweud o'r cyfnodol yn ddweud o'r cyfnodol. Felly, ychydig o'i gofyn am ddweud yma Is it a lot about memory bandwidth that ARM was kind of like invented because there was a problem with using the memory bandwidth or something like that? No, that's sort of backwards. We wanted to use memory bandwidth and we wanted to build certain types of machine and we couldn't do it with what was available. So two things happened to get us out of that. So the first thing was there was going to be a successor to the 6502. So Acorn, like Apple, used the 6502 as its central processor at the time. And we were naturally extremely interested when the original designers, Western Design Centre of the 6502 announced that they were building a successor. So we got an airplane and flew out to Phoenix to see them. Who were those people making that stuff? Western Design Centre. Western Design Centre? Yes. Western Design Centre were a small design house. They sold the microprocessor design on to companies who actually made it. So it was made variously by MOS technology that was part of Commodore and Rockwell and a company called Synotech and several other people. So Western Design Centre were the originators of the processor itself. Was it a sisk? There was another risk? It's insane to characterise the 6502 as any type of processor. If anything, with a very low number of instructions, a very small type, it's difficult to think about it yet at the same time it's clean, not very complex, but it's also not a risk. So anyway, we flew out to see them. Now we've been out to see companies that built microprocessors before. We've been to see national who were the designers of the NS32016. That's definitely a sisk. And that was being designed at various places in the world. We went to see them in Israel. And it was what you would expect. A large industrial unit, chock full of engineers working away on it. They were having a lot of difficulty. Acron was trying to build things using the NS32016. They didn't have anything that worked until Rev D, Rev H, was the first one that was vaguely usable and still wasn't suitable for production. So they were making lots of mistakes. So anyway, we flew off to see Western Design Centre and navigated to the address they'd supplied. And we discovered two bungalows on the outskirts of Phoenix, staffed by some senior design engineers and a bunch of university kids. And they were designing a microprocessor without a large amount of automation, big drawing boards on which they stuck sticky tape. We came away fairly convinced that if those guys were capable of designing a processor, anybody could. So the other thing that happened roughly in parallel with that was the very first papers on teams using the risk paradigm had surfaced. Andy Hopper, now head of the Cambridge University Computing Laboratory, who was at the time a director of Acorn, had put some of those papers on my desk thinking I might be interested to see them. So we had in our hands some pieces of paper showing other people's risk designs. Which year was that? This would be 1983. So you had some papers about risk. And then what happened then? So it's important to remember what risk is. The acronym isn't quite correct in that everybody translated it as reduced instruction set computer. But what it actually is is reduced instruction set complexity computer. So what's been reduced is the complexity of instructions. You don't have instructions that are sort of arcane combinations of things. You simplify that and you simplify it down to the point where in Stanford University's case a small number of post graduates and professors can build one. Ditto at Berkeley and at IBM a small team can produce a microprocessor. So there was a paradigm that seemed to be superior so we started thinking how would you build something. So we played fantasy instruction sets for a while and selected an instruction set and all of that is just thinking. Fantasy instruction set what is that? You select instructions. What would you like your instruction set to look like? If I've put some instructions on the screen I had to want it to look like this that there'd be an instruction that put an immediate value into a register or that moved another register to the program counter. So do you design instructions set or do you choose them? You choose them. It's a choice. So in ARM I made all those choices and with Steve Furber we worked out a way to make it all work. And there are some things that I wanted in the instruction set that he couldn't see a way to making things work and they got left out of the instruction set. Nowadays both of us are much more competent at designing processes and we could have left those in. But at the time we had some guiding principles in the project and one of them was keep it simple. Keep it simple. Yes. So what did you leave out in the beginning? Was it taken back in later? Things that were left out in the beginning set the pattern for how the machine would be. So that's designed of ARM and it stays like that? Yes. So right now there's like billions and billions of ARM devices. So how many of them are very similar to what you did back then? So far ARM have shipped about 50 billion powered ARM chips. So all of them are based on the same what you did back then? There are now about a billion or so that are the very modern ARM instruction set, the thing called ARM V8. I didn't design that. But most of them execute the classic ARM instruction set which I designed the original one of. The majority seller of ARMs is the ARM 7 TDMI. And that essentially runs that instruction set and a compact instruction set that was designed in the early 90s. That's pretty awesome though. How do you feel about the, was it a little bit by chance that it happened? How did you describe it? So the team at Acorn, me, Steve Furber, had rather got used to being able to do everything that we set our minds to. So we must remember that we'd worked together for quite a while building stuff. Building a market processor is essentially building some complex digital logic. And when you had to do that, if you look inside a BBC computer or similar, we're used to building large amounts of complicated digital logic. And that's what a micro processor is. Similarly we're used to programming stuff, the operating system for a machine or the basic interpreter. I'd have written and that's a lot of work, but we're used to that too. So you were making the Acorn machines, right? You were making the software everything or what were you doing? I was doing everything. Were you the CTO? No. Eventually at Acorn I became the system design manager. But it's fairly well known now that I wrote the basic interpreter by myself and designed many of the concepts behind the machines. The operating system of a BBC computer was programmed by Paul Bond and John Zachary, but to my design I said it had to do all of these things and they went off and did it. Well I wrote the basic interpreter. The machine, the BBC computer, is essentially Steve Herbers and my co-design as implemented by the engineers at Acorn, Chris Turner, being the chief engineer, leading them. These machines are huge amounts of collaborative work. If you see a painting by a grandmaster, you know pretty much the grandmaster and maybe some of his helpers did the work on that, but they took them a long time. Engineering isn't like that. It's a team sport. Large teams of people follow in one vision to build something. The vision is obviously refined by everybody in the team to make things better. But something happened in 1983. The papers arrived and from that point until... We're just working in an office every day. What are we doing? How did you get to that point and to a real chip that came a little bit later, two years later? So we designed the instruction set. I designed the instruction set and taught Steve into it. He designed the microarchitecture of the microprocessor. So the microarchitecture is the thing below. The architecture is the instruction set. So we often say, ISA, instruction set architecture. Microarchitecture is the hardware realisation of an engine that can do those instructions. So Steve designed the microarchitecture in parallel with the architecture to prove that he designed the microarchitecture correctly. He wrote, in BBC Basic, a model of the microarchitecture to prove that I designed the architecture correctly. I wrote an interpreter for the processor's instruction set and wrote programmes in it. So well before any actual commitment to doing things, we could demonstrate that you could write sensible programmes in this stuff. The programmes did what I claimed. Steve's microarchitecture did what he claimed. You can then do all the verification that you need to do. So on my models of the processor, you write programmes that are expected to produce particular results. Those models run quite quickly. On a 6502 second processor, we could run an arm emulator at hundreds of thousands of instructions per second where the microarchitecture model and later on the transistor model of the processor ran very much more slowly than that. But we built a test suite that demonstrated the processor was the processor, ran it on my models of the processor and then on Steve's models of the processor to prove that his model did the same as mine. And then when we had a transistor model of the processor, we ran it on that too and then we made one. To nobody's great surprise really, the processor arrived back in April 1985. We plugged it into the development board and it worked. No surprise, we had got rather used to building stuff that worked. We didn't, well we had an idea that we'd done something slightly exceptional because we had the evidence of national semi-conductor not getting the NS3216 quite right for quite a long time. But that it would be a product worthy chip, we didn't quite suspect at the time. So basically it was emulation a bit when you were testing. Is it kind of like emulation when you do it like you said you were doing BBC testing of the system before you actually made the chip? Is it similar to what they do now? Emulation is a specific thing. Emulation you have the detailed RTL model of the processor written in a hardware description language like VHDL or Verilog and you run that on a machine that emulates the effect of all of that. So you have all your RTL, you put it through a silicon compiler or synthesiser and run it on an emulator. So we didn't do any of that. Emulators didn't exist. Emulators came after FPGAs were designed. So there weren't any emulators. What we had were simulators. So Steve's model of the processor in BBC Basic, he also wrote the simulator in there as well. So to model a processor which has clock edges and events of latch data moving, Steve's model of the processor is modeled in a clocked event simulator, a thing that has a regular clock and calls everything. So he'd built a simulator as well as the model of the processor in there. When we built ARM it was built in transistors and extracted on transistor models and we ran the transistors to make sure that the processor functioned correctly. So there weren't any other tools other than those we wrote ourselves and the transistor model of the processor and spice models and so on to check that the transistors worked properly. So maybe we can do this a long time again. Long time ago. You can see some of the acorn machines there. Let's go and check them out.