Most of the discussions are centered around the high cost of migrating to newer technology nodes and the ensuing high cost of equipment, hence consolidation. But the industry is also seeing an evolution of sorts with 3D chips stacking using older technology nodes. So isn't there a potential for new tools albeit lesser expensive that can give a leg up to the smaller companies to stay afloat without getting consolidated?
It seems to me that the talk of consolidation is a bit premature for across the board but will happen in select segments.
Dr. MP Divakar
Its all about streamlining your effort. If they feel lot of energy and money is being wasted in FAB research and they just want to concentrate on the design part what is wrong. I would call it disciple rather than self-imposed strangulation.
Thanks for this article. Yes, you are correct. For years IBM has been a beacon of research in semiconductor technology. I recall reading an article about this project years ago. I hope they are making better progress.
At the moment, the concept of consolidating costs and moving to fab-lite or fabless seems very reasonable. However, I believe that there is a point when the narrowing of options will cause problems for a lot of companies - especially the ones that require the latest and most advanced nodes. As I can see, there are at the moment two advanced options for foundries: TSMC and GlobalFoundries. Samsung will likely become a third option (and could Intel become a fourth? I'm not so sure about that but let's consider it). Suppose there is a company like nVidia. It can not manufacture at Intel nor GlobalFoundries for competitive reasons. As it is pursuing more and more the embedded market with the Tegra lines, maybe it will be hard for nVidia to manufacture its chips at a direct competitor Samsung. So, it practically has only one option - TSMC. If the latter screws a process (and we know how probable this is) it has a serious problem if at least one of its competitors can manage to manufacture to a different facility which does not face these problems.
Increasingly there will be other companies in a similar position.
Why not cut down on R&D as well ... that would save a whole bunch of money too! Sorry but cutting expenses like these doesn't grow your business ... it's a long slow death by self-imposed strangulation.
That's 2 very big 'ifs'. At 22nm and below, second sourcing will be history - transistor variability and ever-more complex DFM will take care of that - and leading-edge capacity, always tight, will be run even more leaner by the foundries (assuming there will be more than one 'real' choice) anxious to run in under rather than over capacity mode. Hey ho!
Peter, that is really an old news - we both know that. IBM is a system and R&D/Engineering services company and usually a decade ahead of the rest of the industry in strategic directions, including with its asset-lite strategy.
Its research is formidable and critical to, for example, GlobalFoundries' long-term viability. I was one of questining TSMC's prowess in research - versus its superb strengths in low cost manufacturing and customer service.
However, recently I was impressed by breakthrough credit to TSMC by Dr. Kelin J Kuhn, Intel Fellow, Director of Advanced Device Technology, Portland Technology Development. She led the development of 90nm, 45nm, 22nm and 15nm process technologies at Intel and is currently engaged in the development of the 11nm CMOS process technologies:
"Recently, TSMC found a method to overcome a technical problem related to fin FETs, which I have been concerning about for a long time. It developed a technology to apply strong strain to the channel of fin FET. This is a major advance toward commercial production of fin FETs."
Of course, we can be sure that both Intel and IBM are deep in the "post-silicon" IC research...
Boris @ Petrov Group
A Book For All Reasons Bernard Cole1 Comment Robert Oshana's recent book "Software Engineering for Embedded Systems (Newnes/Elsevier)," written and edited with Mark Kraeling, is a 'book for all reasons.' At almost 1,200 pages, it ...