At Cambridge Instruments
I worked at Cambridge Instruments as a senior engineer – the title was “Principal Engineer” for what that's worth – from the 8th November 1982 to the 8th November 1991. It the longest period that I worked for anybody, and the work I was doing was relatively demanding and frequently fascinating.
I started off working with a couple of service engineers to integrate a bought-in “voltage contrast” package with an S.100 electron microscope – the idea was to use the electron microscope to make stroboscopic observations of the voltages on the surface of an integrated circuit. The idea had been around for a while, and Graham Plows – the boss of the firm - Lintech - making the add-on – had invented the scheme back in 1968. I've put together a short history voltage contrast here.
This job kept us busy for a couple of months, and got me up to speed for my next job, which was to put together a similar sort of in-house add-on, closely based in work that had been going on at Siemens. The in-house add-on was going to be sold to Thompson-EFCIS in Grenoble who weren't willing to wait until we'd done the job properly. The in-house add-on took another six months or so to get working well enough for us to pass it on to Thompson-EFCIS. In the process we invented a special “beam-blanking system” to go on our column, and got a patent on it.
As soon as we'd delivered the machine Graeme Plows sued us for violating his patents, so we didn't do anything on the project except pay lawyers for the next six months, and I go moved into the S.200 clean-up team. The S.200 electron microscope was an up-graded version of the S.100 electronics which was going to be used to drive the electron microscope column from the S.250 microscope, rather than the cheap and rather nasty S.100 column. Some of the electronic design had been done by less experienced electronic engineers, and it did need some cleaning up.
I then went back to voltage contrast, though we'd resolved the patent problem by sub-contracting the bulk of the development to Lintech, which meant that I concentrated on improving our “beam blanking” system to switch our electron beam on and off even faster than we'd been doing before - ending up with a half-nanosecond long flash.
When that was done I was moved over to electron beam microfabricator work. At that point we'd started selling the EBMF 10.5 – an electron-beam microfabricator that could write at 10MHz across a 5-inch wafer. This was a machine that had been evolving for a decade, and the appearance of the more sensitive electron beam resists that let us write at 10MHz showed up a problem where the thermal noise in our scan amplifiers was making the lines we wrote visibly uneven. I had the job of reworking the scan amplifiers to make them a bit quieter (which wasn't all that difficult) before getting stuck with the job of tidying up the optically isolated 10MHz data transfers from the data processing hardware to the scanning crate, where we faced the double problem that the data transfers were being asked to go faster than the original hardware had been designed to handle, while being transmitted along much longer ribbon cables than the original hardware had been designed to cope with. Fixing that wasn't all that difficult either, but it took a while.
What I hadn't known when the process started, was that Cambridge Instruments was negotiating with Thompson-CSF to commercialise a shaped-beam electron-beam microfabricator. Thompson-CFS believed that what they were selling us was a pre-production prototype, when it was - in reality – a proof-of-principle machine. I was put in charge of the electronic hardware, and we spent six months travelling over to Paris every week to familiarise ourselves with the machine.
By the end of that time, we knew that we would have to dump all of their electronics and start over. It wasn't that their electronics didn't work – it was just that it wouldn't do what our customers would have expected it to do.
We spent about a year designing the electronics that would do that job, while a full-time project planning engineer put together a detailed plan of what we (and the mechanical engineers and the software engineers) were going to have to do and what it would cost. When she'd got results that she was confident about, she predicted that finishing the job would cost some £3.8 million, and tie up every engineer that Cambridge Instruments had for eighteen months.
The money wasn't a problem, but tying up that much engineering effort was. It would have stopped the S.360 electron microscope project, and several others, in their tracks.
Management cancelled the project and spent £3.8 million paying off the people we'd signed contracts with to do the work, including the UK government – the project has been an Alvey project. Some people were made redundant.
I was kicked across to the GaAs crystal-puller division (which had become part of Cambridge Instruments as a result of the ill-fated merger with Metals Research in 1968, and got demerged out again in late in 1988) and spent an interesting year reworking the weighing head in the crystal puller – the original electronics had relied on parts that had become obsolete by 1987. I replace it with a more complicated circuit that the operators liked a whole lot better, mainly because I got rid of a 741 amplifier that was still injecting pop-corn noise in 1987 and replaced it with a more modern part that didn't.
In 1988, Cambridge Instruments took over what was left of Lintech, and hired Graham Plows as it's technical director. Lintech's electron beam tester had been kicked out of the market by Schlumberger's electron beam tester, which had entered the market in 1986 and Lintech hadn't had an order for new machine since then.
Graham thought that Cambridge Instruments should build a better electron beam tester, and do the same to Schlumberger. I knew exactly what he had in mind as a better machine – I'd thought up the idea for that better machine back in 1983 for Ralph Knowles (who had moved to Lintech around 1985) – and Graham got me to write him a brief technical proposal.
I wrote that into my weekly report for the 8th July 1988, and I took away copies of that and every subsequent weekly report I wrote until I was made redundant – with most of the electron beam tester team – on the 8th November 1991.
Graham Plows managed to persuade the Cambridge Instruments executive committee to fund the development of a rather ambitious new machine, based on my technical proposal.
This was a mistake, and if Graham had been better at his job he would have known it. My technical proposal should have been read as “if you are silly enough to ask for 10psec granularity, you can get it but only with expensive new, single-sourced components”. Of course, it wasn't just that choice that doomed us, but also Graham's ambition to have a working prototype at Semicon West in May 1989, less than a year later, which encouraged the project manager – Dave Hall – to skip time-consuming steps like design reviews, and to put everybody under so much pressure that we didn't have time to think about the implications of the corner-cutting we were doing.
That stage of the project is described here. The reports for 1988 can be read here, 1989 here, 1990 here, and 1991 here. The weekly reports won't make much sense if you haven't first read my history of electron beam testing and my somewhat jaundiced review of the way the project was set under way.
Once Semicon West was clearly unattainable, and Dave Hall had moved on to another job, we did manage to turn the proposal into working hardware, and had a machine we could demonstrate from around March 1991. By July 1991 we had three, though it was rare to have more than one working at any given moment. We actually demonstrated one of the machines to Ian Few of Northern Telecom (Canada) on Wednesday 3rdJuly 1991, according to my weekly reports, and there were other demonstrations to potential buyers (though clearly not nearly enough).
This was rather too late – when Graham Plows bailed out, at the end of September 1991, more objective sales personnel decided that we couldn't sell enough machines to make it worthwhile to spend the money that it would take to get them into full production, and the project was cancelled at the beginning of November 1991. Most of the project team was made redundant (including me).