The serious disruptions in the global semiconductor manufacturing and supply chain following the tragic events in Japan triggered painful but productive discussions. Some of the discussions focused on the weaknesses in today's supply chain.
Instead of discouraging us, the natural disasters should prompt us to raise questions such as: Is today's supply chain flexible, thick and adaptive enough to handle future disruptions should they come?
The answer to that is clearly no—and time-to-market, costs, second tape-outs, short lead-times and long qualification times are clearly the main issues.
My humble opinion is that today's supply chain is not only too lean, but should reconsider its basic concepts in regard to flexibility, migration options and most notably second source solutions. Building backup and alternative manufacturing solutions can begin in parallel, in overlap or following the main tapeout with some affordable tradeoffs. Whenever high quantities are involved, these additional NRE costs should be economically justified.
It is almost needless to say that this type of backup tapeout is rarely available.
Therefore, how can a semiconductor company overcome future disruptions? Two guidelines can help:
Fast migration to a second source solution
Moving production to another vendor in another country relies on a series of prerequisites. For a fast migration, predominantly, the design team should have the entire set of required multiple libraries and databases pre-installed and ready for migration. In addition, the design team should be experienced with the additional vendor’s libraries in order to shorten lead times.
In other words, the ability to quickly provide second source chips is now more important than ever.
Conversion to functionally identical components
The required second source chip should be pin-to-pin compatible and functionally identical with the component whose manufacturing has been disrupted. This chip might be critical to your project.
Fast and seamless conversion of chips into functionally identical components, based on an infrastructure of various fab vendors over the globe is an elegant way to provide second source chips.
Yehuda Ben-Yaish is VP of engineering at KaiSemi Ltd., a fabless Israeli vendor that partnered with several tier-one fabs to offer automated FPGA to ASIC and ASIC to ASIC conversions and productization, with guaranteed functionality and pin to pin replacement. Yehuda Ben-Yaish can be reached at firstname.lastname@example.org.
Having a back up for standard off the shelf ICs is pretty easy. The problem occurs when the IC needed is a custom design. It could be an ASIC or a full custom chip. Large companies that have multiple fabs may be able to offer a second source in another fab. This is often an acceptable solution. I just don't see as many customers requiring a second source as they used to.
No one said or think that "These get done in a day", especially someone like me which is in the silicon business for so many years. I know and understand the issues, the costs, the time, the "pain" but still it is doable, exactly as the first one was.
It is not easy, but I still think that "Building backup and alternative manufacturing solutions can begin in parallel, in overlap or following the main tapeout with some affordable trade offs. Whenever high quantities are involved, these additional NRE costs should be economically justified."
I think you should read my post more carefully as I have never said or mentioned another offering of FPGA as a solution. Kaisemi does not offer FPGAs.
I was talking only on ASICs and special ICs, which are ordered by demand and not available regularly on the shelve. Please see our site.
It is completely clear to everyone in this business that this solution is not simple but doable. It is suitable to many required ASICs and probably irrelevant to many others, for many reasons.
The main thing I wanted to say is that specific ASICs (and not only them) which are critical to the supply chain should be available in one of the ways (inventory, parallel FABs, etc.) for a certain recovery time when a serious disruption appears. Today everyone try to minimize or avoid inventory.
Fast Asic-to-Asic conversion is one of the ways for a fast recovery, if nothing else is available, even so not suitable to all ICs.
Ben-Yaish, having design resources in multiple places is hardly the beginning of having a functioning, error-free semiconductor device. With design rules hitting 30nm or below these days, proper process set of and qualification of each location are critical. These do not get done in a day, you know? Too many electrical/design engineers in the semiconductor industry think error-free processes come for free and painless.
You argue that the answer is to continue the fabless model (as lucrative as it's been to a few) with a new backup chain offering FPGAs.
I doubt there will simply be a humongous surge in demand analogous to the loss in supply created by the Japanese earthquake.
While the redundancy in the design of FPGA allows "fast and seamless conversion," (though with obvious and striking differences between the performance of FPGA and hard ICs) "into identical components," high initial cost to logistic redundancy times high initial cost to design redundancy equals a rather expensive "secondary but vital" existence, like some Shaharazadian vizier's dream.
Pre-news: [ http://www.technologyreview.com/computing/37406/?p1=A1&a=f ] I paraphrase:
"The new chips are a cheaper, more powerful competitor. Says Steve Teig, founder and chief technology officer of Tabula, his company's reprogrammable design is considerably smaller than that of an FPGA. 'From its behavior, our [design] is indistinguishable from a stack of chips,' says Teig, making it 5*cheaper to make, providing more than 2*density of logic and roughly 4*performance." [Basically, built-in paging].
Even if you proposed Teig's FPGA instead of your own, there would still be a flawed logistic. If the problem was, at first, that far-off supply was inherently unstable, why build a second chain equally as long and not even always needed?
The fabless, JIT model is showing it's greatest failure right now: the chips aren't here. I understand why anyone would seriously suggest manufacturing more redundancy to be more synchronized, but only if they obviously would stand to profit.
What you should promote is getting out of the fabless model entirely. Obviously, with Texas Instruments acquiring National Semiconductor, there should be no more doubt (at least in America) that "real men have fabs".
-- Gabriel Arthur Petrie
Student of electronics and manufacturing