Cryo-SIMPLY: A Reliable STT-MRAM-Based Smart Material Implication Architecture for In-Memory Computing.

Area of Science:

• Materials Science
• Electrical Engineering
• Computer Engineering

Background:

• Spin-transfer torque magnetic random-access memory (STT-MRAM) is a key technology for logic-in-memory (LIM) architectures.
• Operating LIM at cryogenic temperatures (77 K) offers potential for enhanced performance and energy efficiency.
• Existing SIMPLY (smart material implication) schemes face limitations at cryogenic conditions.

Purpose of the Study:

• To present and evaluate a reliable smart material implication (SIMPLY) scheme operating at cryogenic conditions (77 K).
• To compare the performance of SIMPLY schemes based on single-barrier (SMTJ) and double-barrier (DMTJ) magnetic tunnel junctions at 77 K.
• To assess the energy efficiency of cryogenic SIMPLY schemes for logic-in-memory applications.

Main Methods:

• Development of a temperature-aware macrospin-based Verilog-A compact model for MTJ devices.
• Calibration of a 65 nm commercial process design kit (PDK) down to 77 K using silicon measurements.
• Implementation and simulation of SIMPLY schemes using both SMTJ and DMTJ technologies at 77 K and 300 K.

Main Results:

• The DMTJ-based SIMPLY scheme demonstrated a 2.3x improvement in read margin at 77 K compared to room temperature operation.
• At 77 K, the DMTJ-based SIMPLY scheme achieved approximately 69% energy savings over the SMTJ-based counterpart.
• Cryogenic operation significantly enhances the reliability and energy efficiency of SIMPLY architectures.

Conclusions:

• The SIMPLY scheme operating at cryogenic conditions (77 K) is a viable and promising solution for advanced logic-in-memory (LIM) architectures.
• DMTJ-based SIMPLY offers superior read margin and energy efficiency at cryogenic temperatures.
• This research paves the way for developing high-performance, energy-efficient computing systems leveraging cryogenic technologies.