ARM announces Cortex-A17, the new CPU core that finally replaces Cortex-A9

• By Joel Hruska on February 11, 2014 at 3:01 pm
• • 
  • • ARM has announced new plans for a successor to the Cortex-A12, the upcoming 32-bit CPU core that’s scheduled to ship out later this year, which itself is the successor to the Cortex-A9. The new Cortex-A17 is a further 32-bit evolution of the ARM product family and illustrates something ARM told me last year when I visited its headquarters in Cambridge: The advent of the 64-bit Cortex-A53 and A57 families and the AArch64 instruction set doesn’t mean ARM is finished with the 32-bit market.
      The Cortex-A17 is ARM’s targeted solution to offer high-end 2013 performance in the 2015 midrange market. It’s going to be built on somewhat older 28nm technology (by 2015 standards), which may surprise some folks, but makes sense given the design targets of the SoC. ARM’s own internal cost savings for 20nm match the flat curve we predicted back in 2012, as well as comments from TSMC and GlobalFoundries. With 20nm flat on cost, it makes sense to look for ways to optimize the architecture, particularly in cost-sensitive mobile devices.
      
      With 14nm-XM (GlobalFoundries) and 16nm FinFET (TSMC) providing only small cost advantages on a per-transistor basis, the emphasis at ARM is on building a cost-effective product. The Cortex-A17 is meant to squeeze better performance out of steady transistor counts rather than pushing those counts higher due to cost savings. That dovetails well with mobile’s overall goals, which is to improve performance efficiency in any case.
      Thus, we have the Cortex-A17. The new chip will extend the Cortex-A12′s performance to reach current high-end 32-bit ARM CPU levels, with added support for big.LITTLE, which the current Cortex-A12 implementation lacks. The front-end and execution cores are expected to be similar to the Cortx-A12, with most improvements coming in the memory subsystem. According to ARM’s internal estimates, the new chip will hit a DMIPS/MHz rating of 4.5 compared to the Cortex-A57′s rating of 5.0, the Cortex-A12′s 3.5, and the Cortex-A19′s 2.5.
      
      High-end 2013 performance in midrange 2015 devices may not seem particularly exciting, but as the pace of smartphone performance increases have slowed, we’ve seen longer-lived products and some killer deals from companies like Google. The Nexus 4, Nexus 5, and Motorola G have all been well-regarded smartphones with relatively high-density screens and good battery life, and that’s a trend we expect will continue to strengthen in the next few years. Meanwhile, midrange products have a better chance of penetrating emerging markets, where customers are more sensitive to price and mobile sales are expected to surge.
      When ARM unveiled the Cortex-A12 last year, many people saw it as a reaction to the Cortex-A15′s general set of problems. While the A15 was an unquestionably powerful chip, its high heat dissipation and troubles with big.LITTLE led to weaker market adoption. Qualcomm’s Krait CPU may be Cortex-A15 class, but its power consumption and overall characteristics are superior to the ARM core itself. Nvidia’s own Tegra 4 is Cortex-A15 based, but the T4 hasn’t enjoyed much success in terms of design wins or product shipments. Again, performance wasn’t the problem — thermal dissipation and power consumption were.
      The Cortex-A17 will achieve the A15′s performance, but will do so in a smaller power envelope and with better overall characteristics. If ARM hits the 2015 readiness target, we could see the chip in shipping devices by late in 2015 or early 2016. The delay is the inevitable result of the difference between having chips validated at the foundry and having devices fully integrated and passed through carrier validation.