The foundry battle between TSMC, world's largest foundry and GlobalFoundries, world's second or third foundry – is heating up. Today, TSMC announced that the company managed to clock the ARM Cortex-A9 core all the way to 3.1GHz using 28nm process node.
With 40nm, 32nm and now 28nm process node, foundries such as TSMC and GlobalFoundries utilized test wafers with ARM Cortex-A9 processor logic. The reason is simple – test the ultimate performance of the process node, i.e. achieve as high clock as possible. At 40nm, we saw GlobalFoundries delivering 2GHz, while TSMC delivered 2.5GHz. However, at 28nm process node, GlobalFoundries reached 3GHz first (almost a year ago), while TSMC had its test chips operating at 1.7-2.5GHz. Thus, we were pleasantly surprised when we received a press release from TSMC, stating that the foundry managed to clock ARM Cortex-A9 dual-core processor test-chip all the way to 3.1 GHz.
TSMC used its 28nm HPM (High-Performance Mobile) process node which for example, is used by Qualcomm for its dual-core Snapdragon S4 SoC processors. While you should not expect to see a 3.1GHz processor in a mobile phone, tablets, clamshells and Ultrabooks are whole another ballgame. The foundry now claims the company has now demonstrated its technology leadership in PPA capabilities (Performance, Power, Area). For a comparison, TSMC's 40nm node was nowhere as successful, given that NVIDIA Tegra 3 has five Cortex-A9 cores operating at up to 1.7GHz – the very maximum TSMC was able to achieve with that process node. This was confirmed with the following statement:
"At 3.1 GHz this 28HPM dual-core processor implementation is twice as fast as its counterpart at TSMC 40 nm under the same operating conditions," said Cliff Hou, TSMC Vice President, Research & Development. "This work demonstrates how ARM and TSMC can satisfy high performance market demands. With other implementation options, 28HPM is also highly suited for a wide range of markets that prize performance and power efficiency."
Now you know what to expect with second generation 28nm parts from Qualcomm, and first generation 28nm from NVIDIA, Texas Instruments etc. It will be interesting to see how Intel will compete with its lower clocked 32nm Medfield and upcoming 22nm Silvermont.