IMCRYPTO: An In-Memory Computing Fabric for AES Encryption and Decryption

Abstract

IMCRYPTO is a novel In-Memory Computing (IMC) fabric designed to accelerate AES encryption and decryption using a unified hardware architecture. This approach combines the (Inv)SubBytes and (Inv)MixColumns steps, leveraging high parallelism through RAM and RA/CAM arrays for enhanced efficiency. The architecture achieves exceptional performance gains, improving throughput per area by up to 223.1x compared to existing AES-128 solutions while integrating a RISC-V core for programmability.

Report

Key Highlights

• IMC Acceleration: Proposes IMCRYPTO, an In-Memory Computing fabric specifically designed for high-speed AES encryption and decryption.
• Unified Architecture: Employs a single structure capable of performing both encryption and decryption operations.
• Extreme Efficiency: Achieves significant performance improvements over existing solutions, yielding a throughput per area gain of 3.3x (minimum) up to 223.1x (maximum) for AES-128.
• Programmability: Integration of a RISC-V core provides crucial flexibility and programmability to the hardware fabric.

Technical Details

• Computing Fabric: IMCRYPTO utilizes an In-Memory Computing (IMC) paradigm to accelerate cryptographic operations.
• Structural Optimization: The design combines the traditionally separate (Inv)SubBytes and (Inv)MixColumns steps into a more efficient, combined process.
• Memory Components: High parallelism is achieved through the use of multiple units of Random-Access Memory (RAM) and Random-Access/Content-Addressable Memory (RA/CAM) arrays.
• Future Scaling: Projections indicate that the architecture can gain up to 5.3x benefit in the area-delay-power product when transitioning to emerging technologies.

Implications

• Security Integration in RISC-V: This work demonstrates a highly efficient method for integrating critical cryptographic acceleration directly alongside a flexible RISC-V core, promoting high-performance, secure processing in customizable computing environments.
• Data Security Performance: By achieving unprecedented throughput-per-area improvements, IMCRYPTO makes real-time, pervasive AES encryption and decryption feasible even in power- or area-constrained environments, such as IoT or edge devices.
• Validation of IMC: This paper validates In-Memory Computing as a powerful paradigm not just for AI/NN acceleration, but also for complex, structured operations like cryptography, expanding the scope of IMC applications.
• Custom Hardware Ecosystem: The combination of specialized IMC hardware and a general-purpose RISC-V core provides a strong template for future heterogeneous systems requiring both standardized processing and extreme, application-specific acceleration.