TSMC is building a US$15.7 billion foundry to focus on next-generation 5nm and 3nm nodes

TSMC has announced plans to build a US$15.7 billion foundry to actively focus on next-generation 5nm and 3nm process nodes. While this may seem a bit premature, given that 10nm chips aren’t even out on the market yet, the time needed to get permits to build, actual construction time, and the additional time needed to bring a fabrication plant fully online means that the timing makes sense.

As Moore’s Law approaches its limits and process shrinks become more difficult, the number of companies attempting production has fallen over the years. As ExtremeTech points out, 19 companies had production capacity at 130nm, but there are just five companies at 20/22nm.

As of now, just four foundries offer 14/16nm, namely, Intel, Samsung, TSMC, and GlobalFoundries, with pure-play foundries like TSMC dependent on a small number of customers for the bulk of their revenue. For instance, Apple and Qualcomm combine for over 30 per cent of TSMC’s revenue.

TSMC’s current roadmap has it on track to introduce 7nm process technology by the first quarter of 2018, while Samsung says it will ship 7nm by the end of that year (it recently announced the mass production of 10nm chips). On Intel’s end, the chipmaker has said it will ship its 10nm chips by the second half of 2017, after numerous delays.

However, it’s important to note that this doesn’t quite mean that the likes of TSMC and Samsung are ahead of Intel. Things aren’t as simple as comparing same-sized process nodes between companies, and Intel’s process nodes generally have smaller feature sizes than the other firms. That said, Samsung’s and TSMC’s 7nm node is expected to be more or less comparable to Intel’s 10nm, although that could change further down the road.

TSMC's new fabrication facility will presumably feature extreme ultraviolet lithography (EUV), although it's still an open question as to whether the technology will be ready by then. TSMC's commitment to shrinking silicon also appears to signal a paucity of practical options for moving beyond Moore's Law at this point in time, even as successive die shrinks become even more difficult. 

Source: ExtremeTech

Our articles may contain affiliate links. If you buy through these links, we may earn a small commission.