The crunch will come next year for the Japanese semiconductor industry when it has to decide what to do about 32nm manufacturing.

"We have to make the decision next year," JJ Yamaguchi, executive vice-president and member of the board of NEC Electronics, told EW. "The choices are: Shall we have our own fab? Shall we go to foundry? Or shall we have a joint fab?"

The first option looks the longest shot. "It's unlikely we'll have our own 32nm fab," said Yamaguchi. However, he makes an exception to this for the automotive semiconductor business where customers' quality issues require suppliers to have their own in-house manufacturing.

For this, a relatively small production line, maybe not at the leading edge because of the car industry's multi-year design-in times, will be retained. However, a solely-owned, high volume 32nm fab does not look a likely option because of cost.

"After 45nm, the commitment for fabrication will be very high," said Yamaguchi. The second option, going to foundry, could involve any of the major three companies. "There are many feasibility studies. We know the capabilities of UMC, TSMC and Chartered," said Yamaguchi, "we have not yet decided." Currently NEC makes 90 per cent of its wafers in-house.

The third option, a joint fab, looks the most promising. "Our natural ally would be Toshiba, because we have developed the process technology together," said Yamaguchi.
With 45nm fabs costing between $3.2bn and $3.5bn, according to Intel, the bill for a 32nm fab may be not far short of $4bn.

According to one industry study, a company would need $13bn in revenues to be able to get a return on a 32nm fab, and only Intel, Samsung and, potentially, TSMC, have that.
Earlier this year, Texas Instruments said it will not be building a 32nm fab. As Yamaguchi puts it: "Everyone is in the same boat."