Winbond is a billion-dollar chip manufacturer, mainly of two types of memory: DRAM and serial-NOR flash. In general, memory has not been a good market for several years, with many heavily committed companies taking financial losses on the activity. The return of better average selling prices is certainly encouraging for a lot of memory manufacturers, but for most the respite is only an opportunity for them to decide where their future lies -- in wafer fab manufacturing or in design, packaging, and supply chain management.
The first point that Krishna Shekar, director of strategic marketing at Winbond Electronics Corp., wants to make is that memory is a very fragmented sector with lots of memory and interface types, and so some sectors are up while others are down. For Winbond the serial-NOR and DRAM balance each other, he told me.
"In serial flash we are the No. 1 supplier with 33 percent of the units and 28 percent in terms of revenue," he said. "In DRAM we are not a leader. We are a follower, a couple of generations behind in terms of density and capacity. We do make legacy SDRAMs up to DDR3 and mobile and graphics DRAMs. But we sell for embedded use [on equipment motherboards] rather than try to compete for sales into memory modules."
Krishna Shekar, director of strategic marketing at Winbond.
Winbond manufactures serial flash at 58 nm minimum geometries with a 46 nm process for serial NAND products in development. The company makes its DRAMs on a 65 nm process and plans to go to 40 nm class soon. So it is not at the leading-edge.
Winbond's manufacturing is done at a well established wafer fab in Taichung, Taiwan, that runs 300 mm diameter wafers and has a manufacturing capacity of 36,000 wafer starts per month. But the fab is more or less full and running at close to full utilization, Shekar said. Fab managers are always adept at squeezing extra semiconductor manufacturing equipment into a clean room when there is extra revenue in prospect, but eventually a wafer fab is full and that is that.
So what happens next?
"We have a plan for a second module of equivalent size. So, long-term we could consider expansion or go out to foundry, perhaps another facility outside Taiwan. Customers are interested in the availability of a second source for products. But at this point there is no need."
Assuming the money can be raised for another manufacturing push, the lead times on building and equipping a wafer fab -- 18 months to two years to first silicon -- must surely mean that a decision cannot be put off much longer. Shekar declined to elaborate.