The latest issue of IEEE Solid State Circuits magazine has a good piece by Lynn Conway about how she changed VLSI design. In a quite direct way she changed my career too. In 1979 I was just starting grad school, and working as a research assisstant on a radar signal processing project. I was lost. The math involved was beyond me. As Asimov once noted, everyone has some limit in mathematics, and mine turned out to be maximum likelihood analysis.
But I took a course that fall, “Introduction to VLSI Systems”, taught by Professors John Newkirk and Rob Mathews. It was based on the book of the same name that described an exciting new approach pioneered by Conway and Carver Mead, then of Xerox PARC. They had been looking at this new field of chip design, and trying to find a way to simplify and regularize it. In the mid-70s it was a difficult area, full of obscure and proprietary knowledge.
Their approach entirely clicked with me. I could really see how I could do this. The main feature of the course was to actually design a chip and have it built, and I came up with a nice idea – a self-testing memory. Add some logic to a standard RAM so that it could do its own reads and writes to verify its operation. This is routine now, but quite novel in 1979, since logic design in the old style was so hard.
They also devised an astonishing new way of building these chips. All the students would submit their designs over this new thing called the Arpanet. A server would take each design, check it, and translate it into an input file for an e-beam mask-making machine. All the designs were put onto one set of masks, built at an HP fab, and returned only five weeks later. It was an automated network-centric foundry service, a forerunner of Internet commerce, done in the days of 300-baud modems. This Net stuff was good for something besides email and SF-Lovers Digest.
People went wild with their design approach. Just at my school there were three major innovations in the next couple of years:
- Prof. Jim Clark built the first Geometry Engine, a four-way SIMD floating point unit that is the ancestor of modern GPU chips.
- Prof. John Hennessey built one of the first RISC microprocessors, the MIPS I, which was later spun off into MIPS Technology.
- Prof Abbas El Gamal worked out how to route programmable logic, and became a founder of the FPGA company Actel.
Today it’s known as the Mead-Conway Revolution.
I’ve had occasion to meet Carver Mead in the years since then. He’s been a tremendous figure in the field, but always had time for curious young engineers. I never met Conway, though. She got second billing on the book, and seemed quite retiring compared to the ebullient and charismatic Mead.
It turns out there was a reason for that – she was born a boy. From childhood she had felt like a girl in a boy’s body. She did marry and have two daughters, but it didn’t last. By age 30 in 1968 she couldn’t stand it any more, and started undergoing sex-change operations in Mexico. At that time she was working at IBM. In fact, she had been on one of their advanced programs, the ACS supercomputer. She had devised an elegant means of issuing instructions when their operands were ready instead of just issuing them in sequence This is now known as out-of-order issue, and is used everywhere.
Her brilliance didn’t save her. When IBM management discovered that he was becoming she, they fired her immediately. She changed her name, left her family, and restarted her life from nothing. She worked as a contract programmer, then did processor design at Memorex, and a compressing fax system at Xerox PARC.
The ACS project was a failure, and so was the Memorex computer, and the fax system. She had three strikes against her when she joined the System Science Lab at PARC. Her bosses, W. R. Sutherland and Wes Clark, could see her talents though. Her failures had given her an insight that more successful designers lacked – that knowing how things worked at both the low level of transistors and the high level of processor architecture gave one a synergistic advantage over those who only understood one or the other. If both fields could be simplified enough so that one person could grasp them, great gains were possible.
She first started learning about MOS design in the mid-70s, when Mead was trying to spread the news about this amazing technology within the research community. They worked together on it for several years. They wanted to simplify the layout design rules – the rules that express how close the wires and transistors can be to each other when laid out on a chip – down to the point where they could fit on one page. Only then could students keep enough of them in mind to do a successful design. Likewise they wanted simple rules for sizing transistors to make reliable circuits, and to have a way to describe the layout in a standard way so that it could be built at any foundry.
She didn’t do much more at PARC after the courses and the book came out. She worked on the Strategic Computing Initiative at DARPA in the early 80s, and then became a prof at the University of Michigan. She retired from there in 1999. That was also about the time that Michael Hiltzik, an LA Times reporter, was working on his book about PARC, “Dealers of Lightning”. She came out to a few friends, and then Hiltzik published a big article about her in the LA Times magazine – “Through the Gender Labyrinth”.
It was liberating. She finally married her boyfriend of 15 years, Charlie, and had facial feminization surgery to reduce the Adam’s apple and brow ridge. She has become a spokesperson for trans-gender activism. Her website is packed with material on her own career and trans issues. Her life with Charlie looks to be a pleasant one filled with their hobbies, whitewater rafting and motocross, and pictures of their cats and of their rural place in Michigan.
Did her sexual turmoil contribute to her creativity? It looks that way to me. Getting kicked out of IBM allowed her to connect with the fountain of innovation at PARC. If she had succeeded with ACS, and if they had left her alone about her sex reassignment, she probably would have been shuffled off onto yet another S360 project. Failing at IBM and Memorex was good for her, so long as she learned about what went wrong. Failing at being a man was also good for her, in that she converted physically to the woman that she had been inside. Failing can succeed, so long as it lets you find your true place in the world.