How does a processor work?

13 March 2018

Interview with

Sophie Wilson FREng FRS, Broadcom

The processor is the so-called “brain” of the computer, but what exactly is it doing? Sophie Wilson developed the instruction set that underpins the processors made by ARM, one of the world’s most significant chip design companies - their technology is in about 95% of smartphones. Before that Sophie helped to design the processors that made the Acorn BBC microcomputer possible, and that introduced a whole generation of people to home computing. First off, Chris Smith asked Sophie to explain exactly what a processor is...

Sophie - When we design electronics, it’s all about putting stuff together to make something happen - a fixed function. In the early days, fixed functions were all you got. If you built a radar set for World War II, then it did a fixed thing and it couldn’t vary it at all unless somebody took a big hammer to it and redesigned it. So a programmable element was needed in order to break the German ciphers. Dr Alan Turing put together the very first programmable cypher engines to crack the Enigma cipher and a Turing Bombe was programmable because you had bits of wire on the back that you put into different orders in order to make it run different programmes. Nowadays, we put different instructions into memory so we have a fixed set of functions inside the processor, and it fetches instructions from memory and does each fixed function, and you change the instructions and it does something completely different.

Chris - Those instructions get in through those small connectors on the underside of the chip that Katie was referring to when she built her computer?

Sophie - Kind of, yes. 

Chris - If we were to zoom in with a really powerful microscope on the processor chip, what would we see in there?

Sophie - If you open the chip up, take all the packaging off it, all you’re left with is this coppery coloured ingot of stuff. First you have to etch that away and get rid of it all and then, with a sufficiently good microscope because we make things that are extremely small - critical dimensions on current generation chips are made with deep ultraviolet light so you can’t even resolve this stuff with optical light. But, assuming it’s a sufficiently good microscope - electron microscope or something like that - then you can see lots of layers of connectivity, different types of material. We need to make something capacitive, something connective, something that is a semiconductor, and have all those layers work together. So we build very thin layers of stuff on top of each other to do all this.

Chris - Those are the transistors?

Sophie - Across a silicon chip then we’re making billions of transistors and connecting them together to give the functionality we want. The future of microprocessors is very much the future of transistors and has been for the last 40 years. For some time we’ve had Moore’s Law. Moore’s Law is a law about economics: it says it’s economically feasible to put twice as many transistors onto the same area of silicon every period. The period started off at about a year and then it got lengthened to a year and a half, and now it’s two to three years. Currently, it takes us about 28 times as many scientists to push Moore’s Law forward as it did originally so it’s getting really expensive to do this. We haven’t hit any physical reality limits; we can still do these things, it’s just getting really expensive to do it. So, as I said, we’re using deep ultraviolet light and we want to move to extreme ultraviolet light - 33 nanometre wavelength light.

Chris - Is this so you can etch the silicon to make these tiny components?

Sophie - This is to make the transistors smaller and thus fit more of them in.

Chris - Because if you use light which is a shorter wavelength, then the size of the component you can make is smaller, that’s why you want to use that particular colour?

Sophie - You want to use the smallest controllable bit of light that you can. Making extreme ultraviolet light in sufficient power, because we want about 200 watts of this light, is very hard. I’ve likened it in the past to the Star Wars particle beam weapon so we have a one megawatt carbon laser producing ordinary light. That goes into a vacuum chamber where we have evaporated some tin droplets. We atomise the tin droplets in the vacuum and that produced lots of sets of ultraviolet radiation so we filter out the ones that we want and take those off to be our extreme ultraviolet light source.

Chris - Are there better materials that we can use in future because, obviously, we are getting to the stage now where we are finding these materials are harder to work with to endow them with more power, so is it that we’re just going to step sideways and start using something completely different? Is there going to be a regime shift if you like and we’ll develop the new generation of processors in an entirely new way with a new material?

Sophie - We’ve been using new materials all the way through that. The types of things we use for insulators have changed. How we put the whole thing together has changed enormously. The connectivity has changed - we used to use aluminium, we use copper. In the future we’re going to use really rare things like ruthenium for the interconnect. We call it Silicon Valley. In the future, if we happen to be using molybdenum disulphide…

Chris - It’s not so catchy!

Sophie - It’s not so catchy. Molybdenum disulphide fen or valley - it’s just not going to take off, is it? People have been looking at the future of transistors. We can make things still on a silicon base but with carbon nanotubes, or we can use this molybdenum disulphide material, which is also a semiconductor, and make things smaller but we still have this lithography problem. The people who made the world's smallest one nanometre transistor, they actually made millions of them using carbon nanotubes that they scattered on a surface and selected out the ones that worked, so that’s not really a basis for future mass production. We’re moving towards seven and five nanometre transistors, and when we get to five we really need this extreme ultraviolet laser to work properly. It doesn’t work properly at the moment.

Chris - Just to finish: when you built the BBC microprocessors that went into those first generation of computers that really made a difference to home computing back in the 80s - in the 70s you were designing those weren’t you? How many transistors were on those chips compared to what we’re routinely knocking out for the average smartphone these days?

Sophie - The BBC machine we used an 8-bit microprocessor - the 6502 - and that has 4,000 transistors in it. The very first ARM has 25,000 transistors in it. Currently, if you by a top end GPU…

Chris - That’s graphics processing?

Sophie - Yes. The top end graphics processor, or a top end Intel microprocessor with about 28 cores on it, you’re looking at buying about 9 billion transistors, they will sting you for 10,000 dollars for the best of them. At least it’s practical.


Comments

Add a comment