Blogging from Taiwan: TSMC and 40nm Yield
According to my passport I have been visiting Taiwan regularly for the past 12 years. It has been an amazing experience working with the top semiconductor manufacturers around the world. The density of semiconductor knowledge in Hsinchu Taiwan is staggering, but the past few months in Taiwan have been very difficult.
Starting with the Taiwan DRAM makers and moving to the Semiconductor Foundries, the economic downturn has hit Taiwan hard. In an unprecedented move TSMC publicly laid-off workers and announced that the reduction in force was an “annual performance management development system”. Under this system, a portion of the 5 percent of company employees rated “improvement needed” or “unacceptable” and unable to reach performance targets, are let go. Never in my 25+ year professional life have I seen this type of announcement from Taiwan or any other company outside of the United States! Only in America can an employer publicly degrade people like this, only in America can you label educated, extremely hard working professionals as the bottom of the barrel.
In all the meetings I have attended in Taiwan not once was I the smartest person in the room. Usually I am the funniest, but not the smartest, not even close. The educational standards in Taiwan are very high, you will be hard pressed to find a semiconductor professional in Taiwan without a Masters Degree, and a PhD is common place. So after the layoff fiasco I was not surprised to hear that the displaced TSMC workers and their families surrounded Morris Chang’s (TSMC’s founder and Chairman) house in protest. Nor was I surprised to read about TSMC’s President Rick Tsai’s being quickly replaced by Morris Chang. A recent article in EETimes suggested that dismal 40nm yield was the reason why Morris took back control of TSMC, well don’t believe everything you read.
Today Taiwan is back to its old self. The Taiwan government has consolidated the DRAM companies and both TSMC and UMC are reporting surges in wafer demand and backlogs across industry applications, the exception being automotive of course. But rather than cheer the recovery, analysts claim the surge in orders are due to yield problems which of course are pinned on the foundry. TSMC was very open about 40nm yield challenges starting with the interview with Dr Jack Sun in June of 08.
While working as a Strategic Foundry Relationship consultant, my main focus is yield and design enablement. Once a semiconductor manufacturing process is ramping, yield is generally a factor of design, IP, and/or process. More often than not it is IP, which is why Verification IP (VIP), built-in self test (BIST), and silicon proven (Silicon Aware) IP is an important part of modern semiconductor design. I co-wrote an article last year (Improving Yield with Retooling and Robust Infrastructure) documenting one of my experiences with 65nm yield, which was also a great challenge at the time. Leading foundries spend a significant amount of time working with EDA, IP, and top tier semiconductor companies on yield as it is absolutely a collaborative effort. But of course the foundry is ultimately held accountable for yield and gets the bad press, from EETimes anyway.
Dan,
Stumbled on your blog through EEtimes. Very interesting.
On the issue of yield at foundries, I think that cultural limitations (inherent in the foundry model as deployed in Taiwan) will prove the ultimate roadblock. The highly hierarchical decision making process, the absolute emphasis on cost savings to the point of eliminating common sense, the noise introduced by incessant “evaluations” to either change suppliers or leverage existing ones, all of these things make it very difficult for the average equipment engineer to “do the right thing”. As geometries shrink and the consequences of one’s behavior become increasingly obscured, you get intractable yield issues.
It was once the case that one bought a piece of hardware and could yield provided that minimal standards of cleanliness were applied. We find that newer tools require good raw material, good spare parts and good workmanship to perform consistently well. There are real trade-offs at the high end between quality (or raw material, workers and so on) and yield.
As to the return of Morris Chang, I think it had nothing to do with yield but everything to do with stamping out ill-advised internal practices and provide political cover for what might happen in the future (layoffs and what-not).
New Milford, August 17.