Blog post

Beating Moore’s Law by going faster than time?

By Bill Ray | March 11, 2020 | 0 Comments

We know that Moore’s Law can’t be maintained, stalled by the increasing cost of production – we can’t just make things smaller forever. So, various companies are looking for alternative approaches that might deliver comparable improvements in performance without breaking the bank.

Notable amongst these is Tachyum, a semiconductor startup which is preparing to tape out its first microprocessor for datacenters. The devices are heavy on scale, with the high-end product carrying 128 4GHz cores and 12 DDR5 channels, clocking at 4GHz – we’ll see how that goes when they get the first chips back, but power might be an issue. However, the most interesting features is that the processor is a VLIW architecture, similar to the fading Itanium processor. VLIW works by shifting run time optimization to the compiler, which has a much broader view of instruction sequencing and can, in theory, do a much better job than a narrow-minded processor core. However, Intel took many of Itanium’s tricks and used them to buff up its x86 tools, effectively killing Itanium’s future as the fastest core possible.

Tachyum faces the same challenge of creating a completely new port of all the tools and software needed to make a datacenter hum. But it is 2020, and we’ve learnt a lot about compilers, tools, and execution since Itanium, Transmeta and Chromatic Research.

My colleague Martin Reynolds reckons the company is lightly funded ($25M in its most recent round) considering its ambitions, but if it can demonstrate results as promised it will rapidly attract investment. There’s a lot of money in cloud, and a lot of money for anyone who can maintain the improvements in processor performance we’ve been enjoying for the last fifty years.

For more information about the future of semiconductor design and fabrication take a look at our research note on Semiconductor Technology in 2030.

Leave a Comment