In an interview on the side lines of the ARM Research Summit held in Cambridge this week, he said that if MRAM could achieve a sufficient speed-up to rival SRAM it would be a game-changer for logic design.
Yeric told eeNews Europe that spin-orbit torque MRAM (SOT-MRAM) technology, currently in R&D around the world, might get up to the appropriate speed-endurance trade-offs to allow its introduction into core logic. If it does, it will provide not only the chance to retrofit established ARM architectures with dense non-volatile memories but also to rethink how to design processor cores from the system-level down.
The current generation of MRAM, spin transfer torque MRAM or STT-MRAM, is not really able to do that, although it is starting to be considered as a cache memory option on microcontrollers (see Startup tapes out MRAM-based MCU ). But research institute IMEC has manufactured the follow-on MRAM technology, spin-orbit torque MRAM, on 300mm-diameter wafers. IMEC reported that while STT-MRAM switching speed is limited to 5ns, the SOT-MRAM demonstrated reliable switching down to 210ps (see IMEC makes spin-orbit torque MRAM on 300mm silicon ).
The ability to freeze computing processes on-chip, retain state while drawing no power and then resume, would have considerable consequences, Yeric said. "It would require a new processor architecture. I think we would be adding a new processor line; something that could address a different power envelope in the IoT space; working without batteries by using harvested energy," Yeric told eeNews Europe .
The full interview is available here (see ARM's Greg Yeric on memory, logic and making it ).
Globalfoundries offers embedded MRAM on 22nm FDSOI
ARM backs embedded MRAM on Samsung's FDSOI process
Startup tapes out MRAM-based MCU
IMEC makes spin-orbit torque MRAM on 300mm silicon