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!
With Micron having a serious presence in Taiwan, will they supply memory chiplets to TSM for their silicon on fabric integration? Could layering of memory and processing in the same package be a step towards automata and advanced architectures? Is this even a possibility? Any thoughts of these two companies coming together for radical changes in computing? Are any of TSM's customers looking at something like this? Thanks
With Micron having a serious presence in Taiwan, will they supply memory chiplets to TSM for their silicon on fabric integration? Could layering of memory and processing in the same package be a step towards automata and advanced architectures? Is this even a possibility? Any thoughts of these two companies coming together for radical changes in computing? Are any of TSM's customers looking at something like this? Thanks
Depends on the customer. Do they buy Micron, SK, or SS dram for their HBM stack? You have to remember that memory products are commodity items. The memory maker will have their best practices, the chip designer will buy it, and whomever does the packaging will then stack the dram on top of itself and and the substrate.
Also I might be wrong on this, but Micron's fab in the ROC is a smaller one that they bought years ago. I think their main manufacturing hub is in Malaysia. With small smarterings in the previously mentioned ROC, Japan, and the US.
Will Micron/TSM form an alliance? They did form an alliance to serve their chiplet customers long times ago. TSMC chair’s comments: "I think everyone should not forget who is the technology leader in memory? It is Micron, and Micron has already surpassed Samsung.”
What he didn’t say is far more important than what he said. Given TSMC has a front-row seat in those logic and memory stackings, the chair must have seen many of Micron’s critical design-wins from its chiplet customers. Very likely, he knew why the customers made such a decision.
Memory in the circuit board (commodity market) requires only pin2pin compatible however memory in chiplet needs pad2pad and dice2dice compatible. Chiplet customers, Micron, and TSMC should trust each other and design everything together from the beginning to make it work. Any memory company is either in the loop or out. Once out, it is almost impossible to be back into the loop. The chair’s comments confirmed that Micron has become a critical partner in the whole ecosystem.
One more thing: In this year’s Hsinchu technical forum, one TSMC VP stated that 3DFabric capacity will be expanded to more than 20 times by 2026. With such a massive capacity buildup, one would expect chiplet technology will become mainstream.
