ARM looks set to take its high-end PC and sever chip down to 16 nanometers according to the latest tape-out, which should give it a boost against rival Chipzilla.
If the ARM collective is to compete against Intel in the server and in whatever might remain of the personal computer, then it is going to have to do more than beef up the ARM core so it can take on heavier workloads. Chip foundries that are etching the chips are going to have to close the gap between the process technologies they use to etch their chips and the substantial lead that Intel has when it comes to wafer baking.
Chipzilla has 22 nanometer process technologies well under ramp with the "Ivy Bridge" generation of chips and in future "Haswell" processors later this year, then shifting to 14 nanometer processes for "Broadwell" and "Skylake" over the next several years. Intel already shifted to 3D transistors with its Tri-Gate design, which came out with the 22 nanometer processes and which debuted with the Ivy Bridge laptop and desktop processors.
ARM and fab partner Taiwan Semiconductor Manufacturing Corp, which is at this point the go-to alternative for processors and graphics chips outside of Intel and GlobalFoundries, inked a deal last July to work on tuning up future ARM processors with TSMC processes out beyond the 20 nanometer node.
The plan last year was for TSMC to tune its three-dimensional double-gate field-effect transistors (FinFETs) so they can be implemented with 64-bit ARMv8 processor designs.
The Tri-Gate and FinFET transistor designs have been created to deal with current leakage problems in transistors. As planar transistors keep shrinking as process sizes diminish, they have become relatively poor switches, leaking current even when they are in the off position.
Current leakage is one reason why chips burn so much juice, and you can mitigate the issue by making a big fat fin sticking up, presenting more surface area for the gate and therefore cutting back on leakage.
The good news for ARM enthusiasts is that the company has grabbed the high-end Cortex-A57 chip, which is aimed at PCs, servers, tablets, and other relatively high-end devices, and has taped out the design using TSMC's 16 nanometer FinFET. The tape-out took six months, and was done with ARM's Artisan physical IP and EDA tools from the TSMC and its partners.
ARM Holdings has been demonstrating early instances of its Cortex-A57 chip using 20 nanometer processes, and there is no guarantee that the 16 nanometer FinFET process will be ramped up in time to be what the first 64-bit ARMv8 server chips actually get fabbed in. The first ARMv8 chips aimed at servers are expected later this year from Applied Micro, and early next year from Advanced Micro Devices, Nvidia, Cavium, and others.
ARM and TSMC took six months to tape out a Cortex-A15 using 20 nanometer processes back at the end of 2011, and thus far chips have not been put into production using that older process, either. So there is a lag between theory and practice.
Knowing the issues that wafer bakers have with their most advanced processes, ARM Holdings has been trying to hedge its bets. Last August, the ARM chip steward said it had inked a separate agreement with GlobalFoundries to use its 20 nanometer processes in its Fab 8 plan in Malta, New York, to etch Cortex ARM chips. GlobalFoundries is expected to add FinFET technology sometime after its 20 nanometer node. ®