A little known US-based startup, running on a shoestring budget, has revealed plans to release a 64-core accelerator that, it hopes, will become the third chip sitting alongside the CPU and the GPU, in a bid to help them carry tasks faster on the smartphone itself.
Adapteva, founded by Andreas Olofsson, came up with a multicore architecture called Epiphany which reminds us of Intel's Single-Chip Cloud Computer .
One core is made up of a "high performance" RISC CPU clocked at 1GHz with 32KB distributed local memory and a multicore communication framework and is connected to the other cores in two dimensions.
What's more, Olofsson promises that the chip will operate at only one watt, not uncommon for smartphone system on chips, but significantly less than what you'd expect from a normal accelerator like those found in super computers.
Using a 28nm manufacturing process, Adapteva says that a 250mw, 16-core part would reach around 20GFLOPS while the 64-core part may go up to 100GFLOPS.
The company also plans 256-core, 1024-core and 4096-core; the latter would operate at 64w and reach around 4TFLOPS while occupying 512mm square.
The beauty of Adapteva's architecture lies in its simplicity with linear growth as more cores are added (each of them, running at 1GHz, adding 1GFLOPS).
The company is only three years old and managed to run on $2 million worth of funding, coming up with three working versions of the Epiphany and getting a first customer.
Like ARM and Imagination Technologies, Adapteva wants to license its technology to others, a strategy that may ensure rapid growth in an ultra competitive market.
The biggest challenge though would be to convince system on chip manufacturers like Qualcomm or Texas Instruments to adopt Epiphany, a move that will cost them silicon space on the chip and for which they will have to pay a license.
The good news though is that the ARM-based SoC market has reached a point where there are so many players that it is becoming increasingly difficult to differentiate from each other.
Nvidia & Qualcomm have managed to do a pristine job out of it but what about others? Adapteva and its Epiphany IP may well be a golden opportunity (and a great acquisition target) for one company out there.