TSMC Won’t be Making 20nm GPUs – Investing $16 Billion in Advanced Chip Factory

Author Photo
Feb 9, 2015
Share Tweet Submit

Authentic reports are starting to surface claiming that TSMC will be investing over $16 Billion (a very big amount even for the silicon industry) in advanced fabrication facilities. TSMC has recently been ramping its process up the physics slope and shrinking from 28nm to 20nm and 16nm FinFET. This news is actually very good news for the world of GPUs and SoCs. Because the better TSMC is, the better AMD and Nvidia are.

tsmc_semiconductor_fab14_productionStock photo of a wafer at TSMC. @TSMC Public Domain

Related Rumor: Chipset Diagrams Of AMD’s X390 and X399 Motherboards Leak Out – Hint Towards The Existence of An LGA Socket

TSMC: No 20nm GPUs, Investing $16 Bil in new fab

The report states that a total of $500 Billion Taiwanese dollars will be spent on an the creation of an advanced chip factory (which roughly translates to $16 Billion). However, as most of our readers know, not all investments pan out – the obvious example being 20nm, which has proven to be an absolute thorn for anyone that is not Intel. Last reports suggested that TSMC is only manufacturing low power ASICs on the 20nm node, mostly of the mobility types, the Samsung Exynos and Qualcom Snapdragon chips being the obvious example.

One of the most interesting things mentioned in the report is the fact that it states that TSMC will never make 20nm GPUs. The reason for this is the low yield associated with such big cores and high performance requirements. TSMC will shift to 16nm FinFET in 2016. Surprisingly it wants to transition to 10nm as early as late 2016 – in an attempt to shorten the lead on Intel no doubt. Although with the way things are going in the silicon industry, I wouldn’t put my money on it.

Related All-Inclusive Qualcomm Snapdragon 835 Benchmarks Show Improved Integer, Disappointing Floating Point & Mixed GPU Performance

Intel has already invested around $6 billion in Israeli fabrication facilities and is preparing them for 10nm (cannon lake) node. It has also begun R&D on 7nm. And the transition to that should happen before this decade ends hopefully. The problem of marginal diminishing returns grows ever larger. Exponentially more dollar is required to be invested for ever diminishing returns on IPC gains and node size. This is one of the primary reasons Samsung and GloFo have joined R&D to minimize costs and maximize utility in an attempt to beat the curve. Intel currently has a comfortable gain over the industry but if its not careful, it stands to loose just that.

Share Tweet Submit