If rumours are true, GlobalFoundries may be stealing revenue and customer volume away from major rival TSMC. Back when 28nm was on the drawing board, TSMC’s decision to go with the more difficult gate-last approach was seen as a risky manoeuvre. The Taiwanese foundry also announced it would ramp up multiple 28nm production lines, including options for both conventional polysilicon and high-k metal gate.
GlobalFoundries, in contrast, focused its efforts on gate-first design and went with high-k metal gate across the 28nm process. Between the two companies, TSMC’s bets have paid off more effectively; earlier this year the company claimed to be responsible for 90 per cent of worldwide 28nm production.
There are some signs that this situation might be changing. Recent rumours from DigiTimes’ ubiquitous “sources” imply that TSMC is losing some customer volume to GlobalFoundries. AllWinner, MediaTek, and Qualcomm have all reportedly cut 28nm orders at TSMC due to superior pricing from the firm’s rival.
GlobalFoundries had nothing to say on the accuracy of such rumours, but did point out that it has partnered with several of these companies in the past, while recently publicly inking deals with RockChip and Imec. Samsung may have also snapped up some additional orders.
Rumours on foundry production from Taiwan should always be treated carefully, but there’s reason to think that both Samsung and GlobalFoundries are actively courting customers.
Firstly, there’s the simple fact that GF’s 28nm rollout didn’t go as smoothly as planned and the company will definitely be looking to pick up additional business to fill out its New York foundry. There’s been no word of yield trouble at Samsung, but Apple is reportedly planning to shift some volume to TSMC by next year. Samsung, in turn, has reportedly been prototyping work for other customers, including Nvidia, for months.
There’s been some chatter that TSMC is letting these companies go, having extracted the value of their sales, but I think any serious examination of the foundry business refutes this notion. At the end of 2012, some 58 per cent of TSMC’s revenue came from nodes that were 6-20 years old.
Samsung and GF may be ramping their own 28nm for revenue more gradually than TSMC, and therefore may not see as much of a profit spike in the short term off those lines, but there’s every reason to think both companies could have stable, highly profitable divisions for years to come. GF continues to emphasise its upcoming 20nm and hybridised 14nm-XM FinFET process, but there’s no word yet on when either line will be ready for volume production.