Earlier this week, Rambus confirmed that it is expanding ongoing collaboration with Microsoft researchers to develop prototype systems which optimize memory performance in cryogenic temperatures.
Commenting on the announcement, Jim McGregor, principal analyst and founder of TIRIAS Research, told the EE Times that computing, especially in the data center, is evolving to include intelligence through deep learning, machine learning and AI concepts – all of which require massive amounts of data and compute power.
“In conjunction, the processing architectures are evolving to more advanced computing techniques like quantum computing (which Microsoft is focused on) and neuromorphic computing, like IBM’s TrueNorth processor,” McGregor explained. “These computing resources and AI frameworks will also put more stress on memory subsystems because you have to feed the beast with data to make the system efficient.”
McGregor told the publication that he had discussed the various challenges of cryogenic memory with Rambus scientists, including topics related to materials, operating temperatures and long-term stress.
“[The company] evaluated many different memory technologies and DRAM stood out as being the easiest and least expensive to adapt to the extremely cold temperatures,” he stated.
Rambus, says McGregor, has adopted a pragmatic approach, with the new DRAM utilizing a standard manufacturing process.
“Rambus believes that the various materials will react in a similar manner. In addition, at the lower temperatures, it is possible to reduce the threshold voltage and charge pumps to virtually eliminate leakage current,” he elaborated. “So, theoretically, it should be more efficient and could provide for more performance headroom.”
In addition, says McGregor, the colder temperatures will allow for higher stacks of memory because of the additional heat dissipation. However, he noted, Rambus is not focused on die stacking for the prototypes.
McGregor also commented on the motivation behind cryogenic technology development, stating that “cost and requirements” are playing a major role in driving research and collaboration. Moreover, quantum computing processors also require the advancement of cryogenics.
“So, if you are building a system that requires cryogenic cooling, then it makes sense for the entire system to operate in that environment and to take full advantage of the colder temperatures for system performance and/or efficiency,” he concluded.
Interested in learning more about Rambus’ cryogenic memory initiative? You can check out our cryogenic memory page here.