Martin Banks, Personal Computer World 10/81 - checked

October 1981

'Once upon a time,' Daddy said, as he tucked Junior gently into bed, 'there were a lot of big companies that made a great deal of money making very, very clever things called integrated circuits.

'These were used by other companies that made a great deal of money putting them together into little boxes. Now these did clever things as well and were called microcomputers. The people who worked for these companies were all extremely clever and were able to do things that ordinary people, like you and me, just couldn't do for love nor money.

'Then, there were the people and companies who were just as clever, but in a different way. They were able to sit down and work out all the detailed instructions that were needed to make those microcomputers do what other people wanted them to. It was all terribly involved, and there weren't many people who could do it well. In fact, there weren't that many who could do it at all.'

'Oh, Daddy,' said Junior turning over and closing his eyes, 'I don't believe your boring rotten old story. No-one ever did it like that. Look, I can build a system easy, and I'm only six. It's not quite as boring as Lego.'

How far into the future this particular fairy story has to be set is hard to tell. Probably it will never actually occur in the words as set down. There is a good chance, however, that the idea - that the way in which the semiconductor companies produce chips that others engineer into systems that others provide complex programs for will drift off into antiquity - is quite valid.

What's more, the trends and developments in the design and manufacture of integrated circuits that will cause that drift are starting now.

The first, and possibly most important, is gate array design. The second is the development of firmware modules, which will have a more long-term impact.

The idea of gate array design in integrated circuits has been around for some time. Patents were taken out by Texas Instruments in the early seventies. That company's initial work had been aimed at a technique it called discretionary wiring, where each silicon wafer would hold a variety of different circuit types rather than the normal practice of a large number of circuits that were all the same. The Texas idea was that instead of breaking up the wafer into individual circuits, it would be used intact and that a system function would be produced by wiring together the needed individual circuits. Because of yield problems (where one defective circuit could mean the complete wafer was useless) the technique never caught on. The idea, however, stuck.

It was translated to the individual chips instead of wafer, and instead of attempting individual circuits the idea became that of putting a range of logic functions onto a single chip.

One of the first companies to pioneer this approach was Ferranti with its Uncommitted Logic Array. By 1974 it was producing devices with about 100 uncommitted gates on each and started to corner what was then a very small market.

That market is now set to take off, for the idea of the gate array device is now taking off, and for some very sound reasons.

Gordon Moore, one of the three founders of Intel, is also famous among the semiconductor fraternity as a `law maker'. One of his laws states, in essence, that as device complexity increases, the number and diversity of potential functions that can be incorporated into a circuit also increases. Extrapolate this and a point is reached at which the users of the circuits can legitimately expect to see devices that contain exactly the functions they alone require. But from the industry's point of view there can be nothing worse, for it means that it becomes impossible to manufacture the devices the market wants at any thing like the volume required to make them economically. QED, increasing complexity and high volume manufacture are mutually incompatible.

The gate array neatly sidesteps this problem. By using it, the semiconductor companies can maintain volume manufacture of the basic uncommitted arrays and even gain the inherent advantages of better product yield. The final product will also be able to meet the majority of applications and functional requirements of the majority of users.

This is how it is done. The gate array chip consists of a circuit that is partially manufactured. Each chip contains an array of logic gates and other circuit functions found in every device ever made. They are, however unconnected. It is therefore possible for a circuit designer to carry out the specific task of matching the logic functions to a user's requirements with relative ease. With the design complete the final masks which define the pattern of interconnections are produced and the circuit completed.

The major disadvantage of this approach is that only the majority of applications can be tackled. It should be noted, however, that with the general increase in circuit complexity, that majority is getting ever bigger.

The advantages are that customer specific circuits can be produced quickly, easily and cheaply. It also means that the semiconductor industry can maximise production and process skills and make the best use of the complexity available to it (and the users).

When Gordon Moore first put forward his law, he noted that many of the large systems houses (especially large computer companies) had internal semiconductor fabrication operations making the devices they specifically required. He suggested that the trend would continue.

With the gate array it certainly will and the ability for small systems companies to follow the trend will have some interesting results. Ferranti, for example, states that for many applications the design work is simple enough for the user to do - no more semiconductor 'black art'. This ability of small systems companies to design the circuits they require for the systems products they plan - rather then 'make do' with the devices the semiconductor companies care to manufacture - opens up a whole new growth arena in microcomputer capability.

Specific products advantages (more power, more flexibility, greater communication capability, lower cost and easier manufacture) can now be engineered into new systems as and when required. Just look at what Uncle Clive has achieved with the Sinclair ZX81, which uses a Ferranti gate array, compared to the ZX80. A better computer is also cheaper.

A natural extrapolation of the small microcomputer manufacturer being able to make use of gate array circuits to economically produce exactly what is required is that the user (or more specifically the local distributor/dealer/ systems house) will also be able to do it. Why shouldn't such people, who in theory at least know what their customers want to use a 'silicon foundry', manufacturing other people's designs on contract.

The other development in the integrated circuit business that I mentioned earlier will also have its effect on the microcomputer manufacturers and perhaps to a greater extent the systems houses and distributors/dealers. That development is firmware.

In its simplest form, this is software in a hardware form, and has been around for some time. The read-only memory and programmable ROM are good examples. There are developments here, however, that will greatly extend their influence in system design.

Firstly, there is the area of system utilities. Already there are CRT controllers and keyboard controllers that incorporate both the software and the hardware in a single chip. There are numeric processors that give a hardware implementation of a software floating point maths routine. Intel with its iAPX 432 32-bit processor, is using a kernel operating system in ROM. This trend will continue, and more utilities will start appearing as firmware modules. But why stop there? Why not start putting common program subroutines into ROM? Why not have a module that provides the software linking for these sub-routines? Why not edge towards complete applications packages in the same way?

As the circuit complexity increases, so the memories will get bigger and be able to store more code, so there is no real technical brake. The main one will be marketing - which programs or sub-routines to go for. Another potential brake will be the proliferation of bugs in programs. Put an infested program into ROM and it is stuck - there will be no chance to patch. The whole production run of ROMs would have to be chucked. One way round this to use PROMs instead, for the capacity of these devices is also increasing rapidly. The other alternative is better programming, but. .

Now suppose all this is in existence, what happens then? Well, it would be possible for the systems house/distributor/dealer to configure a system (both in hardware and software) to more nearly suit the exact requirements of a user with simplicity, reliability and low cost. Simplicity, because the task would tend towards the 'pick-a-chip-off-the-rack-and-insert-in-board' approach. Reliability, because semiconductor devices have a high inherent reliability. Low cost, because when made in high volume, the little beasties are usually cheap.

Bear in mind something once said by Gerry Sanders, president of Advanced Micro Devices: 'At the first level of approximation, the semiconductor industry has made memory free. At the second level we have made logic free. Now we are going for the third level, and make software free as well.' The industry can and probably will do it, so think about the implications now.

end