TSMC claims ‘Moores Law is Not Slowing Down’ – Expects 10nm Volume Production by 2017
TSMC is one of the bigger players in the Silicon Industry and is usually the go-to foundry for fab-less design houses with high performance ASIC designs. GPUs fall under this category completely and if what TSMC says is to be believed then we could expect 10nm parts as early as late 2017 or early 2018. However, it is worth mentioning that TSMC is claiming to be able to do what Intel, with all its mountains of cash and decades of expertise, could not.
TSMC: "We don't anticipate Moore's Law is going to slow down anytime soon"
Moore's law is arguably one of the more important trends in computing. While the exact specifics of the definition vary, the generally accepted form is that the number of transistors on a given surface area will roughly double every year or so. While I am sure our readers our well aware of the definition, Intel, which originally came up with the law, has been having serious trouble keeping it alive. It was able to meet Moores law with 14nm process by the skin of its teeth, using broadwell SoCs as a platform to push the node forward.
TSMC on the other hand claims that because it stuck to 20nm planar and went directly to 16nm FinFET, it was able to get the best of both worlds and actually miss entirely the bump that caused Intel to stumble. Theoretically this will allow TSMC to either catch up or even get ahead of Intel. Before anyone runs with the notion, let me be clear, there is zero evidence of this and the only implication comes directly from the Senior Director's mouth, which you can read in detail here. To be honest, I don't see Intel's process advantage shrinking. Whatever time it spends on a node, approximately the same time will have to be spent by the competition as well.
Intel has also stated on occasion that Moore's law is in its last strides, while TSMC claims that they do not anticipate Moore's law slowing down anytime soon. As you can see - that is a very interesting set of contradictions from two of the Industry's most prominent foundries. It is also worth noting that recent reports indicate that TSMC has lost the contract of Qualcomm's Snapdragon 820 to Samsung along with alot of 28nm contracts. If that is the case, then TSMC might have a rocky time ahead of it, financially speaking. Finally, TSMC has also stated that they expect volume production of 10nm to begin by 2017 - which is approximately the same timeframe as Intel give or take a quarter.