ARM, TSMC Tape-out 16nm Cortex-A57 Processor
According to the British mobile chip designer, it took six months for both parties to jointly move from RTL to tape-out. They used ARM's Artisan physical IP, TSMC memory macros, and EDA technologies enabled by TSMC's Open Innovation Platform (OIP) design ecosystem. ARM said this is its highest performing processor yet, ideal for compute intensive applications such as high-end computer, tablet and server products.
The tape-out is the first milestone in their collaboration to jointly optimize the 64-bit ARMv8 processor series on TSMC FinFET process technologies. It follows their previous collaboration in the 64-bit arena with the 20-nm FinFET process technology that began back in 2012.
"This first ARM Cortex-A57 processor implementation paves the way for our mutual customers to leverage the performance and power efficiency of 16nm FinFET technology," said Tom Cronk, executive vice president and general manager, Processor Division, ARM.
The new Cortex-A57 is now ready for mass production, and promises three times the CPU power of current chips like Samsung's Exynos 5, but with the same battery life. It can be be implemented individually, or paired with the Cortex-A53 processor into an ARM big.LITTLE configuration that enables scalable performance and optimal energy-efficiency.
"The joint effort of ARM, TSMC, and TSMC's OIP design ecosystem partners demonstrates the strong commitment to provide industry-leading technology for customer designs to benefit from our latest 64-bit ARMv8 architecture, big.LITTLE processing and ARM POP IP across a wide variety of market segments," Cronk added.
ARM said early adopters can now implement the new 16-nm Corterx-A57 chip into their designs. We'll likely see resulting devices in the latter half of 2013.
Let's see Intel taking the lead in efficiency area first. Then we can discuss whether ARM will be able to catch up again or not...
Let's see Intel taking the lead in efficiency area first. Then we can discuss whether ARM will be able to catch up again or not...
There's more info in the Steamroller CPU thread in the forums.
Cheers!
They already do, according to Anandtech: http://www.anandtech.com/show/6529/busting-the-x86-power-myth-indepth-clover-trail-power-analysis
Now the question is, can Intel get manufacturers interested to switch from ARM?
By the time these chips make it to the consumer, Intel will have already pushed out 14nm products. Judging by the fact that Intel is opening up it's Fab to outside businesses (possibly even Apple), I would assume they will have additional production space to create 14nm SoC's soon after the release of "Broadwell". Intel just got it's feet wet in the smartphone space, but I think it will be a heated battle; Intel vs ARM by 2015.
Tape-out is one of the last steps before before lithography masks are made and first silicon can be produced. Depending on how much of a hurry ARM and TSMC are to have their first test silicon out, they could have their first (potentially) working chips in 3-4 months. The likelihood of nailing it on the first spin is low but it can happen... but even if they do, they will likely need a few more months to tune yields.
If there aren't any major setbacks, it could turn up in consumer devices in late 2014.
They've been at 22nm and will continue with 22nm Haswell this year and 14nm Broadwell in mid/late 2014 or so
In a real production environment, the tape-out is the last step before the design gets shipped out for masks. From there to actual silicon, it takes 2-3 months if all goes well. Since TSMC needs to produce actual silicon to test their process yield, I would bet that TSMC is in just as much of a hurry as ARM is to find out how well the chips will turn out.
There is every reason to believe ARM's tape-out will be made into silicon as quickly as TSMC can accommodate since they both need it: ARM needs to prove their power-efficiency on a process that rivals Intel's and TSMC needs a test device to benchmark their new process yield. Intel usually produces SRAM chips for yield testing - simple regular structure that is very easy to test.
ARM likely isn't expecting fully working silicon on the first spin either. But both companies get something to work from with that... TSMC gets a design to practice/benchmark their yield with and ARM gets silicon to see if back-annotated post-PAR simulation matches silicon results.
Being the first guinea pig on TSMC's 16nm process likely comes with substantial discounts on those first masks and wafers, which is great for early silicon that most likely won't work (or at least not perfectly) on the first try anyway.
For TSMC, it does not matter too much whether or not ARM's design is faulty or not since TSMC can still determine how good their process is by examining the dies with a microscope between metallization layers as they fab the chips.