As we watch the gravitational collapse of the semiconductor industry, it becomes increasingly obvious that the tech zeitgeist, with investment in close lockstep, is squarely centered on complete solutions, not enabling technologies. That this seems unfair (they couldn’t do it without us, and what we do is really, really hard) is irrelevant. The money goes to those who provide value directly to the end-users and no amount of special pleading will change that. Believe me I understand – my roots are in EDA, an enabling technology to an enabling technology. I feel your pain.
Back in the mists of time, Jerry Sanders of AMD said “real men have fabs”, meaning that real semiconductor companies do it all themselves – from design through to delivery of the final product. He probably wishes he never said that. Most semis ceded manufacturing, assembly and test to the foundries and other outsourced experts, allowing them focus on differentiated design expertise – all the way from architecture down to GDSII. A good part of that differentiation is in the back-end of the process, which is very expensive in highly-qualified staff, tooling and contracting with the foundries and OSATs; this represents a major fixed cost for any chip design company.
That may be the cost of doing business if your business is building chips, but what if you are building end-user solutions? You still need the performance, power and cost advantages of semi-custom, but why carry that back-end cost if you only need to build one or two high-performance custom chips a year? Networking companies have seen it this way for a long time. They’ll do all the logic design and rely on an ASIC company to make the rest of the process completely turnkey. Increasingly I think all but the largest OEMs are going to see it the same way, especially now you can buy most of the IP you need. You’ll give up a little in power and performance and maybe some cost, but your competitive advantage is in your complete solution; for your purposes an underlying device may only need to be “good enough”, a position a device maker in a competitive market could not afford.
Which brings me to eSilicon. ASIC as most of us remember it was certainly turnkey, but only after you had navigated a convoluted qualification and estimation stage. To get to a quote could take weeks and tied up a lot of your time. All of which significantly limited comparison shopping. eSilicon has changed this in important ways which are very relevant to this new wave of end-user solution providers, and that starts with how you get a quote. Paul McLellan already blogged about GDSII Explorer, the eSilicon quote generator, so I won’t repeat his detailed summary here. The most important point for my purposes is that you can generate a quote in real-time, on-line by providing just enough information about the process technology and foundry you want to use, die area, package and so on. Importantly, you don’t have to provide a design at this stage. You get NRE and volume costs in just a few minutes and this isn’t an estimate – it’s a binding contract with all the legal boilerplate.
Now imagine how a solution provider can use this. In fact, we don’t have to imagine because Mike Gianfagna (VP Marketing at eSilicon) told me about a real case (no names). Without interacting at all with eSilicon, other than through the website, this company used the generator to build the cost part of their business plan, which they then used to raise funding. Once they had the cash, they ran the generator again to build a final quote, approved the quote and started the engagement. eSilicon had no (human) interaction with them until that point and was able to deliver prototypes in 3 weeks from tapeout. This looks to me like the kind of semi-custom design solution the entrepreneurs of our time really need.
So do real men (or women) use ASIC? If you measure by what you do being really hard and producing the absolute best outcome (for a general market) that it can, then perhaps not. But there are plenty of other industries building high quality complex products which are not known for being highly profitable. If instead you want to measure by who makes the most money, then yes, I think real men (and women) are going to be flocking to ASIC, very possibly built through eSilicon.