Intel – Nguồn thông tin thứ hai được mọi người yêu thích?

A response to Daniel Nenni’s “What’s Wrong with Intel?” article, which invited alternative views.

At the risk of calling down the forecast universal opprobrium, I’m going to disagree with Dan’s take on the centrality of Intel.

I don’t agree that Intel is too big/important to fail or that the US can’t succeed in semiconductors without it. Reading the comments on SemiWiki, suggests I’m in a minority here, but far from alone. Perhaps it’s easier for me to say this, coming from the UK with our widespread tall poppy syndrome (anything big and successful is automatically suspect) and less emotionally invested in Intel.

Nothing here is intended as any criticism of Intel’s people. It just feels now like Intel is fighting forces which can be delayed, but not ultimately resisted. Caught in a pincer between the success of the fabless design model and the relative decline of the x86 business, it arguably can’t sustain leading edge manufacturing without both massive external support and becoming a commercial foundry. But is that what Intel really wants and can succeed at? Broad line customer service was never in their DNA. Why not just split off the foundry side and keep the product group?

In some ways I hope I’m wrong here. A lot of people are working very hard to try to right the ship at Intel, facing the possibly hardest challenge ever seen in the semiconductor business. Much as some of us disagreed with Pat Gelsinger’s IDM2.0 plan and sometimes loose talk, you had to admire his sheer guts and determination. He felt like the last link back to the real men (who always had fabs) of the 70s and 80s – perhaps the last of the IDM true believers. But that world is finally slipping out of view in the rearview mirror.

Companies come and go – they have lifecycles just like their products. The average lifespan of a US company is only 15 years (having dropped from 67 years in the 1920s). At some point, even Intel will wither away and we’ll continue on regardless. Perhaps even CMOS will go the way of TTL, NMOS (as used on the 8086 and 80186) and all those other technologies we barely remember now.

If the US feels in a mess today with all its eggs in the TSMC/Taiwan basket, it is one entirely of its own choosing. It’s tempting to assume here that the choices of US companies and governments over the last three decades were consciously made and that the defense of Taiwan was factored into the cost-benefit analysis of those choices. But apparently not …

So now we are asking Intel to bail out the US by providing a domestic commercial foundry business. Every bit as much as many hope for the opposite – that the US bail out Intel.

Intel built itself – at least over the past 40 years – largely as a high performance microprocessor company. We’re now asking it to become something quite different. Even if that’s possible, I’m not convinced that’s what Intel really wants to – or should – do. It may fit the narrative which demands that Intel serve some vital national security role and start operating as a far more customer service oriented foundry business serving a much wider range of customers and designs. But can you really convert an America’s Cup foiling catamaran into a not quite so fast, but more versatile monohull racing yacht which doesn’t drop off its foils and come to a halt in lighter winds? And in the middle of the race?

What really matters is that technology continues to advance. And from a US perspective, that it retains a leadership position and effective strategic independence in semiconductors (note: I think that’s tolerable from a Western, non-US perspective, since the rest of us have lived with it for around 60 years already).

If we’re demanding that the US have its own foundry company, why not start from a blank sheet? Instead of committing the cardinal engineering error of writing a solution (Intel) into the spec instead of a requirement (we want our own foundry), create a new company. After all, isn’t that what the US does best ? And give the US government a stake if it’s putting up funding – a real, financial stake and not one in micromanaging employment policies. Split the foundry completely off from Intel and get rid of the current conflict of interest with Intel’s product groups. Recall some lessons from the SIA about industry collaboration and pull in talent from other companies. What you end up with may be 80 or 90% from Intel, but it needs to be a fresh start.

It’s often argued that Intel’s product and design groups gain some unique advantage from having close collaboration with Intel’s fabs or that they wouldn’t be competitive without this link. That certainly held in the past, but is far from certain today. Some claim Intel’s design teams are world class and others that they aren’t. Looking from outside with no direct knowledge, it all seems rather confusing and contradictory. Yet we know that for over 40 years Intel have reliably designed and produced some of the most complex, fastest chips seen. We’ve also seen AMD survive and thrive moving from internal fabs to TSMC. So what are we worried about here?

If you think Intel foundry shouldn’t be split off, just remember this: the risk that Intel becomes a follower, second best in everything it does. Intel will be behind TSMC in foundry, nVidia in AI and arguably AMD in x86. Is that what we really want for Intel – to be everyone’s favourite second source?

By all means have a US national foundry champion. Just do it properly. And don’t call it Intel. Let the Intel product group focus and return to its historic excellence. Shoehorning today’s Intel into the IDM 2.0 model won’t help Intel survive. And it won’t ultimately help the US.

I may well be wrong. But however Intel’s future plays out, looking back in 10 or 20 years time, we’ll likely have forgotten today’s chaos and confusion and view the outcome as something that was never in doubt. As Kierkegaard observed, “Life can only be understood backwards; but it must be lived forwards”. Living forwards isn’t going to be easy for Intel for some time. But it can survive. Though perhaps only as separate product and foundry businesses, with only the first called Intel.

Also Read:

What is Wrong with Intel?

3D IC Design Ecosystem Panel at #61DAC

Intel’s Gary Patton Shows the Way to a Systems Foundry #61DAC

Share this post via: