Would TSMC build customer specific fabs?
1 min read
The debate over leading edge manufacturing capacity took an interesting turn recently when TSMC's Morris Chang opened the possibility of the world's largest foundry dedicating fabs to a single customer.
It's likely the debate is focused on Qualcomm, bearing in mind recent comments about the availability of 28nm capacity, but it could also apply to Apple or nVidia. Qualcomm intimated recently that it is dealing with four foundries in order to get the 28nm chips it needs: TSMC, UMC and Samsung have been named, which leaves GlobalFoundries as the fourth.
Dealing with one foundry instead of four would be an attractive option for a fabless company, but would TSMC build a fab for one client? With a significant payment up front to get things moving and sufficient guarantees about where future cash will come from, maybe. The client, meanwhile, would need immense confidence in its future to make a multibillion dollar down payment, commit to taking all the capacity of a leading edge fab and to take that capacity for the long term.
The 28nm capacity shortage happened because a lot of companies decided to go straight to that node, rather than designing at 40nm, where capacity is readily available. High demand, combined with problems bringing up the 28nm process, meant capacity was in short supply.
Sooner or later, there will be enough 28nm wafers to go round and the fuss will die down. But will the same debate take place when the 20nm node starts ramping?
