How to survive in DRAMs

How to survive in DRAMsUnlike others in the DRAM business, Shigeki Matsue of NEC’s semiconductor division can see light at the end of the tunnel and has kept the wheels turning in his UK fab. David Manners reports
The answer to the DRAM problem is simple: choose your customers, ensure they have a regular requirement for advanced product and sign them up for long-term contracts.  
  Recipe for success… Shigeki Matsue, associate senior vice-president at NEC’s semiconductor division. “We’re shipping over 50 per cent of our DRAM output to a selected, stable group of customers who want to make high-end PCs at an early stage”
“We’re shipping over 50 per cent of our DRAM output to a selected, stable group of customers who want to make high-end PCs at an early stage”, says Shigeki Matsue, associate senior vice-president at NEC’s semiconductor division.
That may explain why the wheels at NEC’s semiconductor plants in Livingston Scotland still turn merrily away while those at Siemens’ former DRAM fab in Tyneside and Fujitsu’s fab at Durham are stopped and may never turn again.
NEC is putting ?35m into Livingston at a time when Fujitsu would be prepared to accept ?10m from a purchaser willing to take its Durham plant off its hands (in which some ?600m was invested), and when the government/Siemens taskforce looking for a purchaser for Siemens’ Tyneside fab has all but given up and is disbanding.
But for Matsue the long dark night of the soul for DRAM is coming to an end, and the market’s contraction from $40bn in 95 to $13bn last year will reverse. “The DRAM business will grow more than ten per cent this year because the PC market is not so bad especially in the US and even in Europe it’s fairly good. Only in Japan and Asia is the market flat”, says Matsue.
Unlike many of his rivals in the DRAM business, Matsue is not agonising over whether to make DRAM at all, but over which type of DRAM to make.
“In the fourth quarter Christmas season Intel is expecting 20-25 per cent of the high-end PC will be Rambus,” says Matsue, “if that turns out to be true they will need quite a large number. Maybe, in Q4, 20 per cent or more of our DRAM shipments will be Direct Rambus. NEC is already committed to several key customers.”
In addition to Rambus, NEC has its own proprietary DRAM type called Virtual Channel memory which is licensed to Siemens.
“PC133 Virtual Channel comes between synchronous DRAM and Direct Rambus DRAM in performance, but the cost is very close to synchronous so they will be used in the middle range PC, and in workstations and mainframes where you need a large capacity of main memory”, said matsue, “Direct Rambus is 30 to 40 per cent more expensive per bit so it will be used in PCs where the memory array is not so large. but in mainframe computers or high-end workstations, which have very large memory arrays and where the cost is significant, DDR or Virtual Channel DRAM will be used – and also low-end PCs will use this type, for cost reasons”.
So, by the end of the year, Matsue expects to see 20 per cent of his DRAM output being Direct Rambus and another 20 per cent being DDR/Virtual Channel.
Like Toshiba, NEC is pushing into higher densities as fast as the market allows. “We’ve just started volume production of the 128Mbit, it looks promising, many customers are going to use this, especially the PC and notebook people. We have the 256Mbit and can supply it to people needing a high density block but don’t expect volume production until 2001,” said Matsue.
In process terms, NEC seems conservative. “SiGe (Silicon Germanium) is not for Asia; SOI (Silicon on Insulator) becomes important in high speed logic and NEC is going for a system solution for a mobile phone – but that does not mean going directly to SOI; copper is a problem for a total factory – if we run such a process within a conventional aluminium factory, other devices will be affected. So it needs its own factory. We’ll start to build a copper factory in one year’s time,” said Matsue. As for 300mm, the company will not start to build a factory until 2001.
Unlike many of its competitors in the memory business e.g. Toshiba/IBM in Dominion, Hitachi/Texas Instruments in Twinstar, and the Fujitsu/AMD flash joint venture FASL, NEC does not believe in sharing the costs of a fab. “Theoretically it has no meaning to share a factory 50/50 and share the output 50/50 – the output for the investment is the same”, said Matsue, “so sharing the factory has little meaning but, additionally, it needs lots of negotiation”.

Leave a Reply

Your email address will not be published. Required fields are marked *