Building computing chips is big business. But it takes a lot of investment and brains to stay in the game. In the late 1990s and early 2000s, it was a race to get to the magical 1GHz mark. Then it was 64-bit processing and multicore performance. Now, it’s all about energy efficiency while increasing overall performance, both in instructions per clock (IPC) and in clock speed. For a long time, Intel was the leader in high performance computing manufacturing. However, it’s since been left behind by Samsung. Now, TSMC will have leapfrogged both. It has announcement that the production of chips on the 7 nanometer process has begun.
TSMC’s announcement is quite a big deal. While Samsung has been the process leader with its 10 nanometer process, it’s exclusive to low power, mobile applications. Many in the industry have been hoping that the 7 nanometer process will be a boon. TSMC has now moved from its current 16 nanometer process (most notably for Nvidia’s GeForce 10 series graphics cards) to the 7 nanometer process.
Using deep ultraviolet lithography and argon flouride lasers, this process allows TSMC to build chips on a new process with current gear. However, there have been some alterations in the way the transistors are etched, and the process is still expensive. Jumping from sixteen to seven yields an estimated 60 per cent power efficiency, 70 per cent area reduction per chip and an overall performance increase of 30 per cent.
Improvements to the process are planned by TSMC, with extreme deep ultraviolet lasers planned to be used for the next iteration of the process. This will yield a further power efficiency of 10 per cent and a reduction in chip size of 17 per cent. It’s unknown how much more extra performance that will squeeze out of the same chip design. But judging by the earlier gains, it won’t be more than 10 per cent.
It’s unknown what the first commercial products will be with the TSMC 7 nanometer process. What The Tech understands that AMD plans to use TSMC to manufacture the second generation of Vega graphics cards. Such a process could be a boon for an architecture that takes a lot of power to scale up frequency and performance. Perhaps this is the kick AMD’s graphics division needs to even the performance score against Nvidia.