Superparamagnetic Tunnel Junctions for Logic Devices

Carnegie Mellon University, Pittsburgh PA

Investigators

Abstract

Superparamagnets switch back and forth between two stable states without requiring an external power source, using only thermal energy. This project will investigate the potential for a new type of low power computing using superparamagnets, where the switching frequency can be controlled by a small voltage or current. Nanofabrication techniques will be used to make electrical connections to individual superparamagnets, and then the superparamagnets will be coupled together. To evaluate the potential of this approach for probabilistic computing, logic gates and related devices will be built and tested. for performance and energy efficiency in multiplication and in logic operations. The results will be used to estimate the potential power reduction and processing speed of probabilistic logic gates based on superparamagnetic tunnel junctions. If superparamagnetic tunnel junctions can be optimized for reasonably fast, low power probabilistic computation, the results would have tremendous impact on sensors and hand-held electronic devices, where speed is less critical than battery lifetime. A graduate student will gain experience with nanofabrication and high frequency electronics. Hands-on and web-based demonstrations of logic gates will be developed for high school and middle school students, and both the graduate student and undergraduate researchers will be trained as STEM ambassadors, learning to communicate technical information to a broad audience. The proposed research project will design, fabricate, and test superparamagnetic tunnel junctions that can be controlled by a voltage or current, for use in logic devices for probabilistic computing. Suparamagnetic tunnel junctions will be optimized for large changes in the telegraph signal over relatively small difference in the bias voltage or input current. Different alloys will be investigated to reduce the energy barrier for switching, and therefore the speed of the devices. Hard-wired devices will be fabricated, and the thermal switching rates as a function of bias voltage and input current will be measured using high frequency electronics. Following calibration of the individual superparamagnetic tunnel junctions, they will be coupled together into hybrid circuits and the resulting devices will be tested for use in probabilistic computing. Analog multiplication will involve two independent voltage-controlled superparamagnetic tunnel junctions and a CMOS AND gate. Here the time-average value of the output is predicted to be the product of the time-average values of the input signals. This operation is lower energy with probabilistic computing because it does not require analog-to-digital and digital-to-analog conversion steps. Groups of three interconnected superparamagnetic tunnel junctions will also be investigated as prototype logic gates (AND and OR). The output will be measured as a function of the bias voltages controlling the individual tunnel junctions to optimize the coupling strength between devices necessary for the truth table of the logic gate.