학술논문

A Near-Instantaneous and Non-Invasive Erasure Design Technique to Protect Sensitive Data Stored in Secure SRAMs
Document Type
Conference
Source
ESSCIRC 2021 - IEEE 47th European Solid State Circuits Conference (ESSCIRC) Solid State Circuits Conference (ESSCIRC), ESSCIRC 2021 - IEEE 47th European. :455-458 Sep, 2021
Subject
Communication, Networking and Broadcast Technologies
Components, Circuits, Devices and Systems
Computing and Processing
Fields, Waves and Electromagnetics
Photonics and Electrooptics
Signal Processing and Analysis
Energy consumption
Memory management
Random access memory
Voltage
Switches
Hardware
Foundries
Language
Abstract
On-chip memories, and in particular SRAMs, are among the most critical components in terms of data security because they might contain sensitive data such as secret keys. Whenever a tampering event is detected, one should be able to erase efficiently and rapidly the full content of a memory holding such sensitive data, but current solutions based on simple power-off lead to very long erasure times. In this paper, we present a non-invasive design technique based on an innovative mechanism to remove electric charges from SRAM bitcells still powered on, before refreshing them with a new content not correlated with the previous one. The particularity of this novel hardware countermeasure is to be natively compatible with any SRAM circuit designed from pushed-rule foundry bitcells. We have designed and characterized an 8kB SRAM in 22nm FD-SOI process technology exploiting the proposed security strategy demonstrating an erase operation accomplished in the nanosecond time scale (versus 295µs with the conventional power-off solution) at the cost of an additional area of less than 5%. We have also shown that our solution is more efficient than a solution without prior erasure consisting in writing identical data to all memory addresses in a single clock cycle (1 ns). The use of the latter drops the ratio of zeroized addresses at 92 %, while increasing the operating energy consumption by 2.1x under nominal operating conditions.