You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
Samsung Electronics is struggling with semiconductor yields, endangering its big bets on systems chips and the foundry business.
koreajoongangdaily.joins.com
Kang Moon-soo, head of the foundry market strategy team at Samsung Electronics, acknowledged the difficulties producing chips based on the latest manufacturing techniques during a conference call in January.
"It is true that we have trouble achieving a stabilized yield rate in the beginning stage" of adopting the new manufacturing process, Kang said, citing the complexity of the fabrication process.
That is a bit sugar coated. Yield problems often result in PDK changes and that is a serious design problem. This is all part of being a trusted foundry. Trusted to deliver wafers to meet market windows. Do you remember when Intel went public about shopping for foundry partners? Intel foundry teams were sent to Taiwan and S. Korea. Intel really dodged a bullet when they chose TSMC N3. Had they chosen Samsung 3nm it would have been a massacre with AMD winning big.
That is a bit sugar coated. Yield problems often result in PDK changes and that is a serious design problem. This is all part of being a trusted foundry. Trusted to deliver wafers to meet market windows. Do you remember when Intel went public about shopping for foundry partners? Teams were sent to Taiwan and S. Korea. Intel really dodged a bullet when they chose TSMC N3. Had they chosen Samsung 3nm it would have been a massacre with AMD winning big.
Question on this Daniel - from Intel's perspective - IF both TSMC and Samsung were equal in every respect, shouldn't Intel still choose TSMC to fab when they need extra capacity as Samsung competes with Intel in a lot more areas than TSMC? (i.e. Samsung does manufacture and sell ARM chips for devices that reduce the market demand for Intel chips, while also producing chips for competitors of Intel, while TSMC "only" produces chips for competitors of Intel?)
Question on this Daniel - from Intel's perspective - IF both TSMC and Samsung were equal in every respect, shouldn't Intel still choose TSMC to fab when they need extra capacity as Samsung competes with Intel in a lot more areas than TSMC? (i.e. Samsung does manufacture and sell ARM chips for devices that reduce the market demand for Intel chips, while also producing chips for competitors of Intel, while TSMC "only" produces chips for competitors of Intel?)
Probably Samsung wanted to form TSMC - Apple-alike relationship by making Samsung Exynos team work with Samsung Foundry 4nm. That can be good in some aspects since massive PDK change is allowed. But that will scare other AP customers(like Qualcomm) out. Not sure what's happening inside of Samsung. Maybe their corporate culture is too strict to open yield-PPA related issues early enough? or PDK was too ambitious from the start due to high expectation from executives? Maybe mix of this and that, like what happened in Intel 10nm...
Probably Samsung wanted to form TSMC - Apple-alike relationship by making Samsung Exynos team work with Samsung Foundry 4nm. In some ways, that can be good in some aspects since massive PDK change is allowed. But that will scare other AP customers(like Qualcomm) out. Not sure what's happening inside of Samsung. Maybe their corporate culture is too strict to open yield-PPA related issues early enough? or PDK was too ambitious from the start due to high expectation from executives? Maybe mix of this and that, like what happened in Intel 10nm...
In my experience Samsung foundry culture is to be the first to a new technology at all costs so that is goal #1. Goal #2 is to save your job when goal #1 goes awry.
The process flows are routinely customized on a customer by customer basis. That’s really all semiconductor process engineering is, just tweaking an existing process. It takes commitment from customers though because tweaks take a year to implement; all the tweaks for a year are typically grouped together in one package.
On the leading edge, it’s different, because the process is almost throw-away; 6 months later there will be a new process. There is also very little customer commitment. So it’s a game, no commitment from the Foundry, no commitment from the customer, nobody wins.
Question on this Daniel - from Intel's perspective - IF both TSMC and Samsung were equal in every respect, shouldn't Intel still choose TSMC to fab when they need extra capacity as Samsung competes with Intel in a lot more areas than TSMC? (i.e. Samsung does manufacture and sell ARM chips for devices that reduce the market demand for Intel chips, while also producing chips for competitors of Intel, while TSMC "only" produces chips for competitors of Intel?)
Intel has been a TSMC's customer since the early years of TSMC and when Andy Grove was in charge. The coorporations and relationship are important for both sides.
