Stakes are large in battle for microprocessor market

The Globe and Mail

November 24, 1980

Chip makers are entering their biggest competitive battle yet. This is the fight to see who can dominate the market for the most powerful microprocessor available today - the 16-bit microprocessor that packs the number-crunching ability of a minicomputer onto a tiny chip of silicon.

These are the chips that are going to do the most to transform office routines and industrial processes.

The stakes are large. At the moment, sales of 16-bit microprocessors account for less than 17 per cent of a market worth $800-million (U.S.) a year. By 1990, they are expected to scoop half of a $6-billion market. These chips are expensive - they cost up to 15 times more than eight-bit microprocessors.

They can also require a lot more supporting equipment if they are to be used properly - easily up to 30 support chips, compared with an average of about five for eight-bit microprocessors. However, they offer great advantages.

One is their ability to compute arithmetic sums extremely fast. This makes them particularly suitable for applications in which ultra-fast handling of a number of different variables is desirable - for example, control of industrial processes.

More attractive still are the considerably larger stores of information that these chips can effectively manage. An eight-bit microprocessor can control about 64,000 data items in its memory; a 16-bit device is capable of handling eight million or more.

These are the features that are also expected to make the 16-bit the most important chip in the offices of the future.

Word processors using such chips have larger vocabularies and more text manipulating features. Private, automatic, branch telephone exchanges using them manage more lines while also offering extras like message storing. Desk-top computers can take on tasks normally handled by much larger machines.

The battle for the burgeoning 16-bit market will centre on more than technically sophisticated hardware (the basic circuit structure of the chip). In practice, most customers will need aids to link and use their chips properly. That puts a premium on software (the instructions that tell a chip how to do a particular job) and on other backup support.

Two strategies are possible. One is to manufacture a standard microprocessor and then to provide (cheap) software packages to adapt it to particular functions - programming it like a computer. This, broadly, is the strategy on which two leading contenders in the race - Texas Instruments Inc. of Dallas and Intel Corp. of Santa Clara, Calif. - are betting big money.

The alternative strategy is to design a chip that is custom-made for the individual user - eliminating most of the need for any post- manufacture programming. Having far more experience in it than the U.S. chip makers, European companies could score in this market.

And the Japanese? Happily for the U.S. and European manufacturers alike, the Japanese seem unlikely to be about to repeat in microprocessors the stunning breakthrough they made into the market for memory chips. They lack the necessary software skills. That could change, but not overnight. In microprocessors, the Japanese are not yet challengers.

The U.S. companies dominate the market for standard (as opposed to custom) chips today - not surprisingly. The first microprocessor was developed by Intel's Ted Hoff in 1971 in response to a particular problem.

The then-fledgling company, faced with an enormous order from a Japanese company for its calculator chips, wanted to meet it without becoming overdependent on that one customer, so it came up with a standard chip that could be programmed for different jobs for other customers.

Since then, things have moved quickly. The first microprocessors could handle instructions only in "words" spelled out in groups of four digital letters. These were rapidly incorporated into simple applications like toys and games and traffic light controls.

Before the end of 1971 (with Intel again first off the mark), microprocessors capable of dealing with eight-bit words appeared and were applied to more complex tasks (for example, word processors). In 1976, Texas Instruments came out with the first 16-bit microchip, two years ahead of its rivals.

Does that two-year lead mean that Texas Instruments has swept the stakes on this latest generation of microprocessors? Not at all. The story of how Texas Instruments came to push so rapidly into a 16-bit chip tells why.

When Intel launched the microprocessor revolution with its four-bit chip in 1971, Texas Instruments was caught napping. Its response was to adapt the calculator chips it was then making, coming up with an advance on Intel's chip that it dubbed the four-bit microcomputer. It then swamped the market for four-bit chips - and still dominates it.

However, in concentrating on the four-bit sector, Texas Instruments ignored the rapid development of a market for eight-bit microprocessors. Its failure to come up with an eight-bit processor of its own turned out to be an expensive mistake.

And it led directly to the company's decision to try to stage a comeback by jumping directly into a 16-bit microprocessor, the assumption being that any customer looking for a chip more powerful than a four-bit microprocessor would be delighted to move straight to a (relatively simple) 16-bit one.

That decision meant that Texas Instruments was the first in the field with a 16-bit chip. But the company has paid a price. The new generation of chips has taken longer to catch on than expected, whittling away the advantages of simply being first. Indeed, the market for eight-bit microprocessors has proved remarkably resilient.

Customers have been taking, on average, about two years simply to design 16-bit microprocessors into new, chip-bed products. The design stage does not breed big orders; these flow only when customers are poised to begin to mass produce their newly designed products.

In the chip industry, two years is a long time. By 1978, two years after Texas Instruments had brought out its device, the first rival had appeared - Intel - followed in 1979 by devices from Motorola Inc. of Schaumburg, Ill., and the Zilog subsidiary of Exxon Corp. of New York. These successors, particularly at Motorola and Zilog, are faster, more sophisticated and easier to use than Texas Instruments' 16-bit microprocessor.

Admittedly, the greater popularity of these later arrivals is not immediately apparent from the statistics. In the first quarter of this year, Texas Instruments delivered 130,000 16-bit microprocessors while Intel, its closest rival, delivered just 32,000. But its lead is narrowing rapidly.

The California market research firm Dataquest estimates that in the second quarter Texas Instruments delivered 150,000 chips and Intel 60,000. Because Intel's new 16-bit chips are a lot more expensive than those of Texas Instruments, Intel may already be ahead in dollar sales.

Texas Instruments, of course, has no intention of letting this state of affairs continue. In February, it plans to introduce a new version of its 9900, as its present 16-bit chip is known.

The 9900 lacks speed. Its "architecture" (the way it is structured to solve problems) was designed along much the same lines as that of the company's family of minicomputers - to spare users of the company's minicomputers the expense of adapting their software in order to use it for their Texas Instruments microchips. Grafting an architecture originally designed to suit minicomputers on to microprocessors slowed the chips.

In the new version, the remodeled architecture should put the Texas Instruments product in the same high performance class as its rivals - or so it hopes. Armed with a chip with better hardware, the company thinks it will be more prepared to fight back.

Above all, it feels that it is more prepared than most of its rivals to reduce the software development burdens that 16-bit microprocessor applications impose on chip users. This is important.

The new generation of chips can be programmed faster in easy-to-learn, "high level" languages (using simple, English-like commands). Even so, developing from scratch a set of instructions for a 16-bit microprocessor application is now estimated to cost up to $5-million.

Not all customers have that kind of money. Many of those that do lack the basic programming skills required and will not find it easy to acquire them because of the present world-wide shortage of programmers.

Customers need standardized software building blocks - one module, say, to guide a chip running a display terminal, another to set out a structure for communications among various chip components.

These individual program modules can be connected along with any program the customer may have written - with the help of a program linking module. Intel estimates that such software aids could cut customers' application development costs by as much as 80 per cent.

Intel also concedes that its arch-rival Texas Instruments has at least a six-month lead in developing such standardized software building blocks. Whereas most vendors of microprocessors began as hardware experts - and have had to scramble to develop more software skills - Texas Instruments' long experience with minicomputers has given it a unique familiarity with software.

This could prove to be a decisive advantage. But Intel has strong cards too. First, it has a large user base - having sold more eight-bit microprocessors than anyone else. Many of its eight-bit customers will be happy to stick with Intel in the move to 16-bit microprocessors to avoid the hassle of adapting to a new chip architecture and a new supplier.

Texas Instruments, of course, can similarly hope to coax users of the 9900 to switch to its new version. The problem is there are not many of them. While the company sold 350,000 16-bit micros last year, Intel sold 4.6 million eight-bit micros (plus 72,000 16-bit devices).

Intel's second advantage is that it has many more second sources - licencees (or imitators) manufacturing its product - including Advanced Micro Devices Inc. of Sunnyvale, Calif., Mostek Corp. of Carrollton, Tex., Nippon Electric Co. of Japan and Siemens AG of West Germany.

Texas Instruments has only two for its 9900: the ITT International unit of International Telephone and Telegraph Corp. of New York and American Microsystems Inc. of Santa Clara, Calif. More second sources improve a chip's chance of becoming a world standard.

Intel's third advantage is psychological. It is regarded as the technological leader of the microchip industry. At first blush, the reputation may seem not wholly deserved: Motorola and Zilog have developed the most sophisticated 16-bit chips.

Nevertheless, Intel's announcement of the first 32-bit microprocessor late this year should further enhance its name - if not its sales turnover (32-bit micros are not expected to be big sellers before the early 1990s).

So far, only Intel and Texas Instruments have made clear commitments to spend the massive research and development money that a leading role in the standard 16-bit business demands.

Intel thinks it could spend as much as $200-million developing customer aids "to close the applications technology gap." Here its disadvantage of being small relative to more diversified rivals is offset by the healthy profits that a consistently high rate of innovation has brought.

It is, of course, not wholly a two-horse race. Motorola and Zilog could cream off part of the market. And National Semiconductor Corp. of Santa Clara? National, the laggard among the U.S. micro makers, will be launching its first 16-bit microprocessor only next year.

Its executives shrug off the suggestion that next year may well be too late - insisting that the market for these chips is potentially so large that there will be room for everyone with a good product.

Should European chip makers take an equally sanguine view of the prospects for latecomers in this field? Probably not. Among the potential European contenders, only Philips NV of the Netherlands (through its data processing division) has developed a 16-bit microprocessor of its own. And the Dutch multi-national has been hesitating for months about how to launch it.

Probably, the best bet for Europe's electronics giants - such as General Electric Co. Ltd. of Britain and Siemens - is to take an entirely different tack and to build on their already superior skills in making custom chips.

Custom chips are "dedicated" to particular applications during the manufacturing process itself. They generally do not need programming at all. They are expensive - design and development costs become increasingly heavy as chips grow more complex.

The custom chips of the 1980s will each cram in 100,000 or more logical functions, compared with less than 100 a chip at the beginning of the 1970s. Even now, design and proper testing of the latest chips can take up to 120 man years.

Nevertheless, the case for custom chips is getting stronger every day:

The software building blocks that the leading U.S. makers of 16-bit chips are beginning to offer are only a partial solution to the programming problems that the application of microprocessors presents. Since programming is a skill that is still extremely rare in the development laboratories of most companies, custom chips that eliminate the need for it have great attractions.

The microprocessor itself is generating a demand for custom chips. The basic "families" offered by U.S. manufacturers with their standard micros are only the first tier of the support chips normally needed to make a microprocessor useful. And different ones are needed in, for example, a telephone exchange than in a television set.

As chips grow more powerful, all kinds of products - computers, telephone exchanges, stereo sets - will be built up from a handful of these devices. In such applications, tailor-made chips offer an opportunity to incorporate proprietary features. Many chip users are increasingly prepared to pay extra for them.

Custom chips for special applications in, say, telecommunications or cars can be made in volumes so large that they approach the production runs of general purpose chips. For other applications, they can be made in volumes so small that the costs of their design and development far surpass the costs of their actual manufacture.

At present, the best approach for the European leaders in custom chips lies in a judicious mixture of high and low volume chips.

In future, the need to spread development costs over big production runs may become less acute. Computer-based aids to chip design and testing are already beginning rapidly to drive down costs.

The Europeans have a trump card in their longer experience in custom chips. But they had better work at exploiting their advantage. Right now, the Japanese still have to licence (or copy) U.S. microprocessors. They could soon choose to compensate for this by honing their skills in custom chip making.

That threat would make the U.S. companies sit up. They are aware of the changing economics of the custom chip business. Indeed, much of the most promising research into computerized design and testing is being done in the United States.

If Europe drops back both in general-purpose and in tailor-made chips, the chip makers will not be the only losers. European Economic Community industrial experts, trying to co-ordinate a European comeback in microprocessors, point out that U.S. leadership in the most widely used chip technologies means that the U.S. manufacturers usually also have a three-year lead in developing new chip-based products and processes.