Just last week, we reported that TSMC’s 7nm (non-EUV) process has reached volume production, and that it’s expecting over 50 tape-outs during 2018. TSMC has had a good run of late, scoring some major design wins across a broad spectrum of parts from leading fabless semiconductor companies for CPUs, GPUs, baseband processors, semi-custom chips, field-programmable gate arrays (FPGAs) and application-specific integrated circuits (ASICs).
Today, we have more news from the Far East confirming that several Chinese chip manufacturers, including Cambricon Technologies have tapped TSMC’s 7nm process for their next-gen AI processors.
If the name Cambricon does not appear familiar, that's okay. Cambricon Technologies was only established in 2016, and specializes in designing chips for artificial intelligence workloads. Cambricon recently unveiled their third-generation 1M chip for edge computing applications to power decentralized, real-time, and on-device data processing. Cambricon’s AI tech currently powers the Neural Processing Unit on the Kirin 970 SoC. Cambricon's 1M chip is also designed using TSMC's 7nm process.
Further, given the popularity of cryptocurrency mining, TSMC also stands to benefit from an increased demand for ASICs, which are vastly more efficient at mining, both from a performance and power standpoint, as compared general purpose CPUs and GPUs. In fact, Bitmain, one of the world’s largest Bitcoin mining firms, which also designs its own ASICs, will likely transition from TSMC’s 28nm/16nm processes to their 7nm process in the near future.
What does all of this mean for TSMC? Well, as AI workloads become mainstream and find more end-user applications, TSMC is extremely well-positioned to succeed in an economy that is seeing an unprecedented demand for low-power, high-performance silicon.