Superconducting digital computing (SDC) is an emerging field that promises to continue improving performance in high performance computing (HPC) systems in the coming decades, defying predictions of the decline of Moore’s Law. At Berkeley Lab, our researchers are working to make SDCs operate as efficiently as possible.
SDCs are a class of logic circuits that rely on the unique properties of superconductors, which have zero electrical resistance. Their main advantages over conventional CMOS technology are power efficiency and high operating frequency. Despite their promise, SDC accelerators do not currently provide the multiple orders of magnitude improvements they are capable of because modern design approaches tend to reuse computing architectures (the design of logic gates and circuits) inspired by traditional technologies. These approaches aren’t efficient because superconducting circuits are different: accessing memory is very expensive, but the circuit can operate at hundreds of GHz frequencies.
Our researchers are working on novel computational models based on the representation of information in the time domain to make key operations more efficient, as well as testing and modifying new automated hardware design tools for superconducting electronics.
We develop a fundamentally different design methodology that considers the unique physics environments in which SDC circuits operate. We use Race Logic (RL) which encodes information in the temporal domain, which is more natural to the inherent data representation method in SDC. This approach can provide multiple orders of magnitude efficiency improvements for key HPC applications compared to recycled SDC circuit designs. Contacts: Dilip Vasudevan, George Michelogiannakis
The Intelligence Community (IC) is well known to be a major consumer of high performance computing but is increasingly finding itself frustrated by limitations in overall power consumption and clock speed. The amazing successes of semiconductor technology embodied in Moore’s Law give the impression that computing power might continue on its exponential growth curve indefinitely. However, there are limits of miniaturization and switching speeds imposed by physics as applied to semiconductors, and these limits are now being felt. Clock speeds are starting to stagnate, and device features are now only a few tens of atoms in size, so the search for alternative high-speed and low-power technologies must move on to more exotic materials and design concepts Superconducting Electronics (SCE) offers a promising alternative to complementary metal-oxide semiconductor (CMOS) technology. However, as with many disruptive technologies, in order to displace the reigning champion, there is a lot of ground to make up. New pulse-based logic families operating at very low power levels are starting to be developed, but if they are to compete with semiconductors, they will have to show performance advantages for highly complex circuits. The semiconductor industry has had the advantage of decades of development of ever more sophisticated design tools that keep creating ever more sophisticated circuits. Contact: John Shalf