Cryptocurrency mining hardware designer KnCMiner is planning to deploy its next generation bitcoin ASICs in early 2015.
The Swedish company announced that its upcoming Solar ASIC will be manufactured on the new 16nm FinFET node, allowing for significant performance and efficiency gains.
In a statement, the company said that it would be relying on solutions developed by its engineering team, noting:
"The 16nm process represents a step change in processing capability as we're using several proprietary techniques developed in-house to reach speeds an order of magnitude above today's levels."
KnCMiner unveils specs for new ASIC
KnCMiner became the first bitcoin ASIC designer to launch a planar 20nm chip earlier this year. The first Neptune chips were manufactured a few months ago and the transition to the new process yielded a significant performance improvement over previous generation 28nm chips. The 20nm Neptune features 1440 cores and consumes 0.57W per GH/s.
The new Solar platform is expected to offer a six-fold performance increase compared to KnC's previous generation 20nm hardware. The gains on the efficiency front are even bigger, as the company said the new Solar chips will achieve 0.07W per GH/s.
KnC's Solar ASIC features more than 5,000 cores, making it substantially more complex than the Neptune. The 16nm process enables higher densities than the 20nm node, allowing designers to integrate more transistors per square millimetre.
The new manufacturing process also enables higher clocks and superior efficiency, but one can only speculate on the exact figures at this point. KnCMiner does not disclose the names of its manufacturing partners, making it impossible to draw direct comparisons between nodes.
Bitcoin ASICs go 3D
KnCMiner's announcement represents yet another step toward the manufacture of next-generation bitcoin ASIC chip processes amid a broader shift among the chip industry as a whole.
FinFET allows designers to utilise non-planar transistor designs, therefore chips based on the technology are often said to feature 3D transistors. Different chipmakers have different definitions of what constitutes FinFET manufacturing processes, but the aim of all these new processes is the same it has always been – to deliver better efficiency and superior performance.
Intel was the first chipmaker to start employing non-planar 3D transistors in commercially available chips, but these chips are referred to as tri-gate designs rather than FinFET designs. The technological distinction is blurred – although Intel does not use FinFET branding, its 14nm and 22nm chips, namely Ivy Bridge and Haswell generation designs, can informally be described as FinFET chips.
While Intel has been making inroads in the foundry business in recent years, the company does not lease its latest manufacturing processes to third parties. As a result, chip designers have a choice of Globalfoundries/Samsung and Taiwan Semiconductor Manufacturing Company (TSMC) processes.
TSMC's 16nm FinFET node (16FF) has already entered risk production and the company described yields as excellent. The company pulled in its 16FF process by roughly a quarter and now expects to commence volume production in the first quarter of 2015. TSMC was originally planning to start volume production in the second quarter.
Samsung and Globalfoundries joined forces for their FinFET push earlier this year. The two companies said they should be ready to manufacture their first 14nm FinFET products by the end of 2014, but the ramp-up comes later, sometime in 2015.
Few potential downsides
One of a handful of downsides of shifting to 16nm FinFET, and indeed every new node, is the potential for low yields. Mature processes do not struggle with yield issues, and these issues tend to take a much bigger toll on complex, large chips such as high-end GPUs. Yield issues mean that manufacturers simply get more faulty dies per wafer, pushing the unit price of healthy dies up.
However, this problem is usually faced when dealing with high-volume chips, as poor yields can quickly erode the chipmaker's margins. Bitcoin ASICs are a low-volume affair with a very short lifecycle, so any potential issues will most likely be outweighed by superior performance. Additionally, TSMC has made previous claims that its 16nm FinFET yields are good.
Chips built using the latest manufacturing processes also tend to cost somewhat more than those built on mature nodes, but once again the price premium is outweighed by superior performance, even in consumer chips, let alone bitcoin ASICs.
Solar brand artwork courtesy of KnCMiner