The show must go on: a reliability assessment platform for resistive random access memory crossbars

Area of Science:

• Materials Science
• Computer Engineering
• Artificial Intelligence

Background:

• Resistive random-access memory (ReRAM) is promising for computing-in-memory (CIM) architectures in machine learning (ML).
• Current ReRAM technologies face challenges including cell variability, read disturb, and limited endurance.
• Existing characterization platforms lack comprehensive software stacks for easy system integration and CIM operation support.

Purpose of the Study:

• To introduce the NeuroBreakoutBoard (NBB), a versatile, integrable, and portable instrumentation platform for ReRAM crossbars.
• To enable comprehensive characterization of ReRAM devices and facilitate CIM operations.
• To provide a user-friendly software stack for experiments via Python.

Main Methods:

• Development of the NeuroBreakoutBoard (NBB) platform.
• Implementation of a software stack for Python-based experiments.
• Case study using TiN/Ti/HfO2/TiN ReRAM cells for diverse experiments.

Main Results:

• The NBB platform successfully characterized individual ReRAM cells.
• NBB demonstrated the capability to perform CIM operations.
• Experiments showed a relative measurement error below 2% for CIM operations.

Conclusions:

• The NeuroBreakoutBoard (NBB) is an effective platform for ReRAM crossbar characterization and CIM operations.
• NBB addresses limitations of existing platforms by offering a comprehensive software stack and portability.
• The platform facilitates advancements in ReRAM-based computing for machine learning applications.