1 November 2011
(Odometer: 7,832 miles)
LEXINGTON, Mass.–Andreas Olofsson cut his chip-designing teeth working on Analog Devices’ TigerSharc DSP. He learned that bigger isn’t always better, and he also found out he wanted to build something of his own.
So when circumstances changed, he gathered up his retirement savings, a handful of engineers and set up shop next to one of American history’s most famous byways: Battle Road.
It’s an unlikely spot to put a chip startup. Silicon Valley, yes. Nearby Route 128, sure. But right next to where British soldiers staggered back toward Boston in 1775 after being routed at Lexington/Concord? Unlikely.
But here Adapteva is. Olofsson and team have designed what they believe is a hyper-energy-efficient processor architecture, many-core Epiphany chip, an accelerator for next-generation DSP tasks such as speech recognition and imaging processing. The initial implementation packs 16 custom floating-point cores in a 65nm chip that deliver 50 GFLOPS consuming less than a watt. (See Startup aims many core chip at smartphones.)
He’s not going after the big, established players in the market, but instead feels there’s a market for co-processors to help as media-heavy applications take off (think face and voice recognition among many others). It’s an FPGA co-processor strategy in many ways, and the programmable devices’ flexibility likely means Adapteva will find itself in many different applications.
BittWare is among Adapteva’s customers and is an investor in the company.
Olofsson talks about his startup, his philosophy of design and why bigger isn’t necessarily better: