Tuesday, September 30, 2014


Indians were aware only of the IT software industry, which employed thousands of Indian programmers and paid them well. But behind the software industry was a hardware industry, whose challenges were different. I reviewed them in this article in Business Standard of 13 November 1999.


All that we see in India is the high-flying software industry; we think that the sky is its limit, and reward its companies with enormous valuations. But how does it look from inside? It is easier to get a close look sitting here in Silicon Valley.
Behind the software industry is a hardware industry; without hardware, software would not exist, and the shape of hardware determines the design of the software for it. But no two industries could be more different. The building blocks of hardware are integrated circuits; they have been getting more and more complex, and differentiated. And as they increase in complexity, their process technology becomes trickier. The result is that if someone wanted to set up a fabrication facility – fab as they call it here – it is unlikely to cost less than $2 billion.
And it is not as if you invest $2 billion and then keep producing the chips for the next thirty years. The technology of ICs is being updated every day, and customers want nothing but the latest. So as soon as you have found the couple of billion and built a fab, the struggle starts – to turn out the latest custom-made chip for the yuppie customer. The pressure is not only to turn out tomorrow’s product today; it has to be made at a fraction of yesterday’s price. As Bruce Goldman reported, the price of storage disks was $3 a megabite not long ago; today it is 2 cents, soon it will fall below a cent. And in the meanwhile, storage density goes on increasing, and chips pack more and more memory.
So you can be sure that if you make today’s product and try to sell it next month, the market will have disappeared. Speed is the core of the matter. Computer manufacturers want deliver within hours; they may vary their order by 50 per cent, and you will have to bear it. And although the chips keep changing constantly, a computer manufacturer will expect absolutely standardized, uniform quality. He does not want to worry whether the chip came from one fab or another; they all have to give exactly the same performance.
This means that once the process has been set up for making one chip, it must be left completely unchanged; no tinkering with it. This is all right as long as the chip is coming out of one fab. But suppose you have a success, the demand soars, and you have to build another fab. The only way of making sure that it will give the identical chip is that it must be an exact copy of the first fab. Engineers find it heart-rending to build an identical fab; they would have thought up lots of improvements since the last fab was built and would want to incorporate them in the new one. But that is banned; improvements are too risky if they change the performance of the chip.
Fabs are costly; to be competitive, they have to be built and run cheaply. Take Komag, which has two fabs close to San Francisco. It has another four in Malaysia, one in Japan and one in Thailand. Three-quarters of its production comes from Malaysia, which has the lowest labour costs. Why not India? Too much bureaucratic hassle; a fab would take years to build, not weeks. The product cycle has come down to nine months; India could never permit such fast reaction times.
Designing, manufacturing and selling chips has become a risky business; companies have been looking for less hazardous strategies. One is to buy chips. Many chip manufacturers source chips from other makers if the orders do not fit into its production schedule. Cisco makes internet equipment; it contracts out PCBs to half a dozen manufacturers across the world. And contracting out has created another type of manufacturer – one that does not design chips at all, but makes chips on designs given to it by its customers. This is what Taiwan Semiconductor Manufacturing Corporation has done; as a result, it controls over a third of global foundry work. It has been such a success that it is today the third most valuable semiconductor company in the world, behind Intel and Texas Instruments. The converse of TSMC is a company that only designs chips, but does not manufacture them.
Still, hitherto there has been one constant in this constantly changing industry – the giant manufacturer-suppliers of chips for standard PCs and workstations. They stand between the computer manufacturers at one end and the fabs at the other. The mass manufacturers of computers do not yet go directly to the fabs. Maybe that will happen next. If it does, the giants of the hardware industry, those market valuations have zoomed up and up for a decade, may well be doomed.

So what can they do? They have to run their vast operations with the same nimbleness, the same flexibility as the small internet companies or software designers. Therein lies the riddle – a riddle which so many American companies, and not only the ones in the information technology industry, are trying to unravel. GE owes its success to the solution it has developed – setting its engineers the clear goal of getting close to customers and helping them make better profits year after year, and letting them alone as long as they are performing this core function well. Something similar is happening to the automobile industries. Car manufacturers are letting individual factories design and market their own models, encouraging them to tailor the models to the customers’ requirements, introducing flexibility in an industry characterized by assembly lines. The problem is the same in all engineering industries – where investment in plant and equipment is high but the products demanded by the market are changing. The IC industry is an extreme example: the pace of change there is many times faster, hence organizational solutions are that much more urgent. What this industry does today, other industries will be doing tomorrow.