Kash, questions and comments on your post:
1. Yes, I do think that for some things 5 -6 engineers can get together and start a fabless and turn out a product e.g. DVD, playstation MIPs, and perhaps even fibre-channel, if there are no patent restrictions -
case 1 no patent restrictions like DVD - when market matures, LSI's competition is TSMC/UMC. But they will make both the margins for foundry and also for the fabless guys who produced the design. I don't see that as a problem for the next yr and a half. The market is not matured yet for DVD. This model will work for simple chips, e.g. DVD. How about a CPU core + graphic + hardware MPEG2 + DVD + cable modem + digital TV tuner on one chip at 0.12 um? Or automotive engine management + GSM + wireless Internet access? Or even a simple x86 - just look at all the money AMD/NSM/IDTI is losing Q in Q out. On products like those, there need to be strong coupling between design and processing. That's hard to achieve with a 5 - 6 guys and $2M and foundry fab. But the time we get to 0.18um/0.12um, products like these are the norm, not the exceptions.
case 2 where there are patent restrictions or marketing barriers. I am not sure fibre-channel falls into that or not. But even here, since this is more high-end, the design-in times and stuff will make it hard to crack the market. For example, everybody knows Qualcom buys flash or serial EE or this or that, but there is no way anybody can reasonable expect to crack that account with a startup unless it's such a specialized chip that nobody has.
2. In the long run,perhaps after the next upturn, LSI's competition will also be TSMC and UMC, assuming the latter can amass enough standard cells that can work together. Yes, they will be licencing a lot of these standard cells from companies like yours, designed by very experienced engineers. LSI's advantage would be cells designed by less experienced engineers but all under one roof and perhaps better communications between people designing the different cells. This reminds me of the DRAM start up who wants to make a business out of licencing their design because they made their design portable and scalable from one process to another. Which model will win? Or perhaps both will survive? IMHO, I see chioas in trying to tie 10 modules from ten different design houses together and not only make it functional but also try to jack up the speed grade by one grade every Q.
To summarize, I can't really see which model will prevail in 5 year. I personally take it with a grain of salt that some stock analyst can do that either. As for the next yr to 2 yrs, when I expect the upturn to come, I still see LSI as very nicely positioned with the products that they have. I do agree though that they need to grow by acquistion or sell out to STM. To me VLSI does not have enough mass to survive long term. VLSI should just merge into LSI - great synergy, great cost reductions by eliminating a lot of redundancies, use VLSI cash to pay for Symbios, fill Gresham quick etc. But I guess all these are just pipe dreams, too much egos in the way.
patrick |
