Sohmers started an exascale computer systems company called Rex Computing at the age that most teens are finding it impossible to get out of bed.
His company has just been given $1.25 million to hire another few engineers to create his Neo chips from concept to production—and has a sizeable DARPA contract.
Sohmers has locked down the architecture and should get a verified RTL by the end of this year. It will sample its first chips in the middle of 2016 and move to full production silicon in mid-2017 using TSMC’s 28 nanometer process.
The DARPA funding fixed the automatic scratch pad memory tools which was the “difficult part and where this approach might succeed where others have failed is the static compilation analysis technology at runtime.”
Automated step at runtime is one of the biggies in the Neo design. It will mean you can have a cache-based system but without all the area and power overhead.
Sohmers’ cunning plan is to strip all the complexity out of the on-processor memory system and put it into the compiler. This happens at compile time and does not add to software’s complexity.
The compiler knows where data will need to be at different points and it inserts it rather than storing it in DRAM.
Sohmers’ team has already created the instruction set architecture and the basic core chip design and for the rest of the year will push the functional verification. This will ensure the ISA is optimized for the applications Sohmers’ team are targeting and free from logical inconsistencies and other potential problems.
Sheesh, kids of today. When I was his age I was writing Pascal programs to create a game where a Roman legionary threw a spear at a charging Celt.