US government plans for a post-Moore's Law future

Too early to pick a winner in the battle to replace silicon

The US government needs to increase its support of new paradigms including quantum and neuromorphic computing if it is to remain a leader in the field of high-performance computing.

That's the conclusion of the National US Strategic Computing Initiative (NSCI) in a strategic plan issued recently.

The NCSI is a cross-departmental body with the objective of sustaining and enhancing US leadership in high-performance computing (HPC) into the middle of the century. Its executive council includes members of the departments of Defense, Health and Energy as well as from the FBI, NASA and the National Science Foundation, among others.

The strategic aims of the NCSI include accelerating the delivery of exascale computers (devices capable of a billion billion calculations per second, 100 times more powerful than the current 'standard' 10 petaflop supercomputers), pulling together traditionally siloed modelling and analytics capabilities, and disseminating the advantages conferred by the research and development it supports throughout government and business.

This strategy is complicated by the imminent demise of Moore's Law - the observation that the number of transistors on an integrated circuit doubles around every two years. The physical properties of silicon-based chips mean that they cannot be shrunk indefinitely and the Semiconductor Industry Association, among others, believe this physical limit will be reached within five years.

"Increasing performance beyond exascale or further reductions in energy cost may not be feasible with current semiconductor technologies," the NSCI notes.

"To address these challenges, NSCI activities must explore and create new highly scalable, programmable, power efficient, and economically viable computing technologies."

The NSCI foresees a "more heterogeneous future computing environment, where digital (von Neumann based) computing is augmented by systems implementing alternative computing paradigms to efficiently solve specific classes of problems."

It will continue to support traditional chip development while at the same time looking actively at what might take its place, the plan says.

"The NSCI will pursue two parallel lines of effort over a 10-20 year period: the R&D of technologies that will move digital computing performance past the theoretical limits of complementary metal oxide semiconductors (CMOS) and the research and development of alternative paradigms that will open up new possibilities for the advancement of large - scale computing."

The NCSI will increase the levels of R&D funding and other support in promising areas including quantum and neuromorphic computing as well as related technologies such as superconductors and nanotechnology. However, it says it is still too early to pick a winner among the new computing paradigms.

The US faces severe challenges in its quest to remain the dominant power in HPC over the coming decades. Earlier this year China took the top spot in the supercomputer rankings for the first time with the Sunway TaihuLight, a 93 petaflop machine featuring microprocessors designed and made in China. There is also now a larger number of supercomputers in use in China than the US, again for the first time.