I recommend "Sinclair and the 'Sunrise' Technology - The Deconstruction
of a Myth" by Ian Adamson and Richard Kennedy, pub. Penguin 1986.
Sinclair allowed the authors to interview his PR department and also
myself, an employee of Sinclair Research Ltd. It was the only occasion
when an outsider was allowed to interview an employee of Sinclair
Research. (Also their article in New Scientist 12june1986, on the same
Kennedy spent three or four hours in my house, and was coughing badly.
He had a good grasp of the Sinclair syndrome. He was conscious of lost
opportunities in the computer industry. As a youngish journalist, he had
less grasp than me, but was still impressive. Please tell me if you turn
either of them up.
The latest article on the structure of the super-computer that we should
have built but will not build is mine in Electronics World March 1989,
"The Kernel Machine". It has a million processors. Also see Sunday Times
12mar89, sect. D p14.
Sinclair falls in love with WSI.
Extract from a very accurate account of the Sinclair Syndrome;
Sinclair and the 'Sunrise' Technology
By Ian Adamson and Richard Kennedy
pub. Penguin 1986, pp 225-231.
Given that if the researchers get it right, their work is certain to revolutionize the international microelectronics industry, it seems worthwhile devoting the remainder of this chapter to an explanation of what a wafer is and why it's so important.
Wafer-scale technology came into Sinclair's life on 3 July 1983, when Ivor Catt responded to the advert in the Observer in which Sir Clive announced that he was seeking personnel for his revolutionary MetaLab. In terms of the new centre's stated objectives, Catt was made to measure.
A maverick computer theorist with too much integrity and experience to be tethered to the short-term commercial objectives of an intensely conservative industry, Catt was anathema to the multinationals. For twenty years he had developed the theoretical foundations for a revolution in the semiconductor industry, and for twenty years his ideas had been rejected as unworkable. When Sinclair made it clear that he was interested in Catt's proposals, the disillusioned engineer saw his chance to show the computing Mafia the error of its ways.
By the time Catt had made his bid for the MetaLab, an abortive attempt at wafer-scale integration had become almost a rite of passage for the multinationals of the industry. Everyone who was anyone had had a stab at it. Both TI and ITT had floundered in spectacular style, but it was Gene Amdahl's bid for the super-computer-via-wafer-scale that proved to be the kiss of death for research as far as investors were concerned. Before going into detail about the Amdahl disaster, let's race through a schematic overview of what the technology is all about.
Ivor Catt first applied his mind to the problem of wafer-scale-integration when he identified a significant wastage in the semiconductor manufacturing process:
I noticed that the silicon wafer was a hundredth of the cost of the total system so why not use that cheap commodity to build the system on the wafer instead of sawing it up to form separate circuits (Infomatics, August 1984)
In other words, one of the most conspicuous weaknesses of semiconductor manufacture is the manner in which it has elected to deal with the rejects. Chips are produced on 4- or 5- inch wafers of silicon, which are then chopped up in order to separate the working chips from a large number of rejects. The consequence of this process is that the working chips must then be mounted, wired and packaged into plastic before being assembled on to a printed circuit board (PCB). The wafer-scale concept seeks to do away with this costly and cumbersome second stage development process.
The essence of the wafer-scale alternative is that the entire wafer is preserved (including the reject chips), and designed and tested in such a way as to obviate the need for a PCB. The process requires not the creation of a wafer containing perfect chips, but a method of testing and rejecting that results equally in a sequence of working chips to be built on the wafer. Faulty chips are rejected by electronic logic built into the wafer rather than by cutting the wafer apart.
A variety of approaches have been developed in an attempt to realize such a process, but the Catt solution can be summarized as follows. One of a number of possible entry points is selected on the wafer, and the first chip in the sequence is tested. An off-wafer tester feeds random data to the chip, and if the appropriate data is returned to the tester the first chip is considered to be sound. The next stage in the process requires the tester to instruct the working chip to open a link with one of four adjacent chips. Data is once again sent to the second chip via the original working chip, and if the appropriate data is returned then the second chip is considered sound. If the second chip fails to return the appropriate data, then the last working chip is instructed to open a link with one of its three remaining adjacent chips, and the test sequence is repeated. In this way, a spiral of working chips is established, in the end creating a one-dimensional array of chips that function as what is known as a 'shift-register'.
Now you don't have to know anything about one-dimensional arrays or shift registers to get an intimation of the beauty of such a process. The inherent economies of the process lie in the fact that the rejects can remain in place and that the working chips are formed in a working configuration that doesn't need to be reassembled on to a separate PCB.
As far as investors in the US were concerned, Gene Amdahl was the golden boy of the sunrise industries. Having made his name as one of IBM's foremost hardware designers, Amdahl had moved on to set up his own company, Trilogy……
…… Trilogy found no shortage of willing investors, US heavyweights Sperry and Digital Equipment came up with the backing, as did the Bull Corporation in France. At conservative estimates, more than $240m. were pumped into the Amahl's project.
….The only party with cause to gloat over Amdahl's failure was IBM, who in the 1960s had decided that wafer scale integration was too difficult and costly a development to be a viable proposition.
In the light of the experience at Amdahl, IBM, and Texas Instruments, Catt's optimism seemed unfounded and his association with Sinclair Research insignificant. However, as far as Catt was concerned, the opposition's failure to realize wafer scale was no more surprising than would be the failure of the wheel without the concept of the circle. According to Catt, Amdahl's inability to bring wafer scale to fruition was a result of a failure to recognize the advantages of the technology his company was attempting to develop. Axiomatic to the Catt approach is the exploitation of the reduction in interconnections facilitated by wafer-scale integration. It's generally believed that in the latter stages of its development, an Amdahl mega chip boasted an astounding 1200 pins packed on to its 2.5 inch length. As a consequence of ther testing process outlined above, Catt's wafer chip requires only two pins, since communication with the component is necessarily confined to the first chip in the spiral. Equally, entire chips are rejected only if each one of the first chips at all possible entry points reveal a fault.
The uniformly negative evidence seemed to suggest that the only hope for wafer-scale integration lay in the creation of fault-free wafers, the production of which, according to Gene Amdahl, will take more than a century to perfect. Unless, of course, one attached any significance to the theories of a crank like Ivor Catt.
To his credit, Sir Clive didn't dismiss Catt as a crank, but he identified with the inventor's dedication to a project that had been rejected by the industry's establishment for two decades. Sinclair's support for Catt's vision is multiply determined. In the first instance, having been shunned and scorned for his perennial devotion to electric vehicles and pocked television, Sinclair would have recognised and valorized the drive of a kindred spirit. Less emotively, Sir Clive's consuming interest in developments that reduce production costs by limiting the component count would have stimulated his interest in Catt's theories. (It is no coincidence that the ZX81 was the first product to take advantage of Ferranti's revolutionary development of ULA technology.) Finally, the logic and simplicity of Catt's theories are so seductive that only the vested interests and hidebound conservatism of the multinationals could find reason to deny their experimental implementation.
It should be stressed that while the computer industry shunned Catt's theories, over the years the inventor's approach to wafer-scale development has found support from both the state and the academic world. The NRDC funded research that enabled Catt to develop his concepts to the stage where he was able to patent its implications, and at the Middlesex Polytechnic Dr Malcolm Wilkinson [incorrect. Actually it was R Aubusson. ICjan99] headed a team that explored the practical implementation of wafer-scale a la Catt. Over the years, the combined assault on the industry by Catt and Wilkinson have generated a positive response, but in the event ICL, GEC, Plessey and STC all turned down the project.
While Ivor Catt eked out a living as a lecturer at Watford College, Wilkinson managed to persuade Burroughs [now UNISYS. ICjan99] to take him on as head of a team investigating wafer-scale's potential. The team's research reached the stage where the theoretical basis of the project was demonstrable, but further resources were required to examine the commercial viability of the process. Wilkinson's group had produced a working wafer when the project fell foul of company politics:
The Burroughs wafer was very much a test structure. Exactly the time when the wafer project was being viewed at Burroughs, very major structural changes were taking place in the company. There were a lot of people at intermediate levels who were very sensitive to these changes and thus not prepared to take risks of champion risky technical problems or projects. And at that time the wafer was considered to be fairly high risk, particularly since a lot of people who were coming into Burroughs were from IBM and were quite hostile to wafers. (Interview with Malcolm Wilkinson, 15 December 1985.)
In September 1983, Sinclair and Catt had a series of meetings at which the implications of wafer-scale development were discussed at length. As a result of these discussions, Sir Clive bought the rights to Catt's patents, and took him on as a consultant. In February 1984, Malcolm Wilkinson left Burroughs for Sinclair Research and was soon joined by three others from his original team. The wafer-scale project soon became the centerpiece of MetaLab research. In its heyday it boasted twelve full-time members headhunted from Plessey, Texas Instruments, STL and GEC.
By March 1985, the wafer team had made sufficient progress to convince Sir Clive that the fruits of its endeavour might well constitute the foundations on which Sinclair Research could re-establish itself as a market leader. The problem was, wafer-scale development took the company into the new area of component manufacture in which it had no track record and for which it required significant investment. To drum up financial support for his new endeavour, Sir Clive recruited the expertise and clout of old friend and ICL chairman Robb Wilmot.
Wilmot's original brief was to seek out £50m. worth of investment for the creation of a wafer-scale production plant. As far as the financial press was concerned, Wilmot's arrival ensured that attaining the venture capital was merely a formality. In the event, the financial crisis culminating in the abortive Maxwell rescue chronicled above proved too much even for the ICL chairman. As far as the city was concerned, the combination of a Sinclair product and semiconductor technology amounted to a less than tempting package. However, although Wilmot severed his relationship with Sinclair Research in September 1985 without finalising the investment package required for the wafer-scale plant, it seems that his contribution to the project was far from negligible. According to Wilkinson, it was Wilmot who recognised the potential the wafer-scale development.
I think that there's a lot of money to be made out of the wafer project, perhaps a lot more than Clive originally envisaged. I think it was Wilmot who eventually showed that the potential of all this is much larger than its effects on Sinclair Research. We could end up with a large semiconductor plant servicing a much wider range of the marketplace than Sinclair Research could handle … Clive had simply looked at the wafer concept and considered how it would impact Sinclair products. I think what Wilmot did was to take the marketplace in general and ask how wafer scale would impact mainframes, telecommunications and military systems. (Interview, 15 December 1985).
While the announcement was lost in the collapse of the Maxwell rescue, the most compelling motivation for investment in the new Sinclair venture is that Wilkinson's team has succeeded where the multinationals have failed. In May 1985, the group was poised to go ahead with the production of a half-megabyte wafer-scale memory for the QL. Then a slump in the memory market meant that standard RAM was being sold for less than cost price, and the Sinclair product was no longer economically viable.
At the beginning of 1986, Sinclair's wafer-scale enterprise still appeared to be the company's most promising hope for the future. Negotiations initiated by Wilmot seemed close to fruition, although at the time it seemed unlikely that Sinclair Research was in any position to achieve a controlling interest in any such enterprise. The situation was further complicated by the fact that Catt's patents were in the hands of Barclays, part of the security covering the overdraft that was keeping Sir Clive's company afloat. So although salvation appeared to be at hand, it was still possible that the company might flounder at the final hurdle. Malcolm Wilkinson put the situation into perspective:
It's difficult in the context of Sinclair Research today to see how they can realize the benefits of the resource [wafer scale] quickly enough. They really need a product for this Christmas , and the wafer is a long-term development. (ibid).
Now that Sugar has removed the financial drain and distractions of the troubled parent company, and Barclays Bank has come up with the necessary funds to kickstart Anamartic, it's up to Wilkinson and his team to prove that they really can take the world by storm. One thing's for certain: the rewards that would accompany a successful realization of the product would make Sinclair Research's heyday look like a depression. And as a significant shareholder in the company, clearly Sir Clive, must have in mind precisely the kind of renaissance that success with wafer would bring.