|From: Don Green||4/9/2009 1:42:28 AM|
|What Could Microsoft Want With a Chip Guy?|
It’s not too surprising that microprocessor guru Marc Tremblay has decided to leave Sun Microsystems, which was experiencing challenges and executive departures well before the brouhaha over stalled takeover talks with IBM. More intriguing is the fact that he is going to Microsoft, which is not exactly a center of chip design.
Tremblay, in an email, referred questions to a spokeswoman for Microsoft. She could only provide a statement with a few boiler-plate facts about his new job: He will hold the title of distinguished engineer in the “strategic software/silicon architectures” group under Craig Mundie, Microsoft’s chief research and strategy officer.
This is not a group that many people knew existed. The spokeswoman could not answer when it began operating, or how many people are in it. But she said Tremblay will manage a team of technologists “who will help set the company strategy for software and semiconductor technologies, as well as maintain relationships with semiconductor companies.”
Stepping back, it’s easy to see how a person with Tremblay’s talents could help the company. Microsoft’s Xbox division, for example, has to think about which microprocessors to consider in designing a follow-up to its current gaming console. Its Windows group, meanwhile, has to design new versions of the operating system for the rapid proliferation of chips with many electronic brains rather than one or two.
Tremblay, who was chief technology officer of Sun’s chip unit, certainly has the credentials. During 18 years at Sun, he amassed at least 100 patents–the most of anyone at Sun–and led the development of several important members of a chip line called Sparc that has long powered Sun’s flagship server systems.
That hardware represents a sliver of the market compared with machines based on x86 chips, the kind sold by Intel and Advanced Micro Devices. But Sun in recent years put out an eight-processor Sparc chip–part of a line that had the code name Niagara–that has sold very well for small servers.
Tremblay, whose departure was reported Tuesday by the New York Times, is more closely associated with a chip called Rock that was designed for high-end machines. And Rock has not been such a happy story; in February, Tremblay told reporters that the chip, which will have 16 processors, won’t be ready until the second half of 2009–compared to an original arrival date of the second half of 2008.
And that part of Sun’s server line faces long-term questions, whether or not IBM decides to buy the company. Billings for those systems declined 32% to $662 million in the second quarter ended in December, while the Niagara-type machines grew 31% to $369 million.
|RecommendKeepReplyMark as Last Read|
|From: Arthur Tang||4/9/2009 5:25:06 AM|
|There is a 'titanic' shift in semiconductor designs that Microsoft is interested. Mainly, surface cmos circuits or buried cmos circuits. The speed is night and day in favour of surface circuits produced by ion deposition with much less linewidth and layer thickness, avoiding wet chemicals. The buried circuits fill up holes, and use wet chemicals to do chemical mechanical copper polish(highK gates). Which can not be more precise in structure tolerances. So, the cpus are limited to 3.8 ghz. Surface cmos circuits are being developed upto 75 ghz target.|
Further more, the data processing of multicores can be 8 bit(ascii data streams) instead of batched branch prediction data processing of 32 bit multicores. 4 word batched processing(21 stages) is slow.
For the next few years, we are going to see ultra low core voltages(microvolts or nanovolts) to make computers without too much heat or using too much electrical power.
So be it. But then, Java may want to sell to Microsoft instead of IBM?
|RecommendKeepReplyMark as Last ReadRead Replies (1)|
|To: Manly Dirk who wrote (64808)||4/10/2009 4:59:03 AM|
|From: Arthur Tang|
|I have been in the semi industry since 1958. Knew the Intel people years ago personally, Dr Bob Noyes, Andy Grove and Craig Barrett.|
I designed a wafer scale computer in 1980, when yield was only 90%. So research was done on dry chemicals to achieve 100% yield. We used ion(graphite) deposition to put a coating of diamond on Yag. Ion deposition equipment is basically 5000 volts from source to target in Argon atmosphere. Layer thickness is very precise and as thin as you want.
We did chemical mechanical polish, and thickness is not precise. So, gate dielectric thickness control is poor for higher cpu clock frequency. Intel semi process has a frequency barrier for many years now. Java has its cpus made by Hitachi. They also suffer frequency barrier in their semi processing technology. So, the whole industry needs to be cautious beyond the two camps of technology, IBM and Taiwan semi.
New semi industry plants in China will all go with ion deposition equipment.The technology of linewidth, 2 nm was achieved by Intel Hillsboro, and 3 nm at MIT Lincoln lab in 1998. The production of 32 nm is still being developed. Ion deposition made 2 nm linewidth and layer thickness easy to manufacture.
|RecommendKeepReplyMark as Last Read|
|From: Arthur Tang||4/11/2009 5:22:11 AM|
|You may be interested in the future of computer design?|
Gene Amdahl was interested in my wafer scale computer concept. He was the designer of S370 vector mainframe computer for IBM. His wafer scale design was a few 4 inch silicon wafers at Amdahl in 1980, published in Scientific American. Mine was 8 inch with 100 8 bit cpus, multicored to be switchable to 800 bit mainframe computer tied together with shift registers.
You either process data with 100 channels of 8 bit data or 1 channel of 800 bit data or any divisions there of.
Of course, today, we go for ultra low core voltage to reduce heat and run very high frequency clocks(75 ghz target). Some production is being done in China for their future data center in each city.
|RecommendKeepReplyMark as Last ReadRead Replies (2)|