SERAD: Soft Error Resilient Asynchronous Design Using a Bundled Data Protocol

Sai Aparna Aketi, Smriti Gupta, Huimei Cheng, Joycee Mekie, Peter A. Beerel

TL;DR

SERAD is a novel asynchronous hardware design template that protects chips from radiation-induced "soft errors" without the massive area and power overhead of traditional methods. By using a "bundled-data" asynchronous protocol, the system only incurs a delay penalty when an error is actually detected, allowing it to maintain near-baseline performance during normal operation.

What Problem Does This Paper Address?

Electronic components in space or military environments are bombarded by high-energy particles. This causes two main types of "soft errors":

  1. Single Event Transients (SETs): Temporary voltage glitches in combinational logic.
  2. Single Event Upsets (SEUs): Bit-flips in memory elements (latches/flip-flops).

Traditional Radiation Hardening by Design (RHBD) techniques are "pessimistic." For example, Triple Modular Redundancy (TMR) runs three copies of the logic and votes on the result, tripling the area and power. Glitch Filtering (GF) adds delays to every path to "drown out" glitches, which permanently slows down the clock frequency. The authors identify a gap: we need a design that is resilient but doesn't sacrifice performance and power during the 99.9% of the time when no radiation strike is occurring.

Methods

The authors introduce SERAD (Soft Error Resilient Asynchronous Design). Its architecture relies on several key pillars:

  • Asynchronous Bundled-Data Protocol: Unlike synchronous chips that use a global clock, SERAD uses local handshaking signals. This flexibility allows the "clock" to simply wait if an error needs to be corrected.
  • Error Detecting Logic (EDL): Each pipeline stage has a transition detector. If a glitch (SET) appears at the input of a latch while it's trying to capture data, the EDL flags it.
  • Temporal Redundancy (Re-sampling): When the EDL detects an error, the SERAD controller stalls the pipeline, waits for the glitch to pass, and re-samples the data. This "temporal" approach avoids needing three copies of the logic.
  • DICE Latches: To handle SEUs (bit-flips), the design uses Dual Interlocked Storage Cells (DICE), which are inherently resistant to state changes from single-node strikes.
  • Hardened Control Logic: The "brain" of the system (the asynchronous controller) is hardened using Guarded Dual Modular Redundancy (GDMR) and specific gate sizing to ensure the control signals themselves are never corrupted by radiation.

Datasets

The researchers evaluated SERAD using:

  • Benchmarks: The "Plasma" open-source MIPS-like processor core.
  • Workloads: Standard embedded benchmarks including "count" (arithmetic), "pi" (calculation), and "CoreMark" (general performance).
  • Simulation Tools: Spice (for analog/transient accuracy) and Verilog (for digital logic validation) using the NCSU 45nm cell library.

Key Insights and Ideas

  • The Asynchronous Advantage: The "aha moment" is realizing that asynchronous design is the perfect host for error resilience. In a synchronous system, stopping the clock to fix a glitch is a global nightmare; in an asynchronous system, it's just a local delay.
  • Performance "Pay-as-You-Go": Because SETs are relatively rare, the average performance of SERAD is nearly identical to a standard, unhardened chip. You only "pay" the time penalty when a particle actually hits the chip.
  • Counter-Intuitive Power Savings: Surprisingly, the SERAD version of the processor sometimes consumed less power than the unhardened synchronous baseline. This is because the transition from flip-flops to a latch-based asynchronous design removed many power-hungry local clock buffers and reduced unnecessary switching activity (glitches) in the sequential elements.

Key Results

The authors compared SERAD against TMR and Glitch Filtering (GF) on the MIPS-like processor:

  • Area: SERAD uses less than half the area of TMR (80% overhead vs. ~300% for TMR).
  • Performance: SERAD achieved a maximum frequency of 333 MHz, significantly faster than the Glitch Filtering approach (286 MHz) and nearly matching the unhardened baseline (340 MHz).
  • Power: In the "count" benchmark, SERAD showed a 23.7% reduction in total power compared to the unhardened synchronous baseline, despite having more transistors.
  • Resilience: Analog simulations confirmed that SETs with widths up to the design's defined threshold were successfully detected and filtered without causing logic errors.

Limitations and Future Work

  • Leakage Power: While dynamic power is low, the increased transistor count (from DICE and EDL) leads to higher leakage current (about 5.9x higher than baseline). This could be a concern for ultra-low-power standby applications.
  • Physical Design Complexity: Asynchronous designs are notoriously harder to implement with standard industry "Place and Route" tools compared to synchronous designs. The authors note that clock tree synthesis for these designs remains an ongoing engineering challenge.
  • Future Optimization: The authors suggest that the "transparency window" of the latches could be tuned to detect not just radiation errors but also standard timing violations, potentially allowing the chip to run even faster by "borrowing" time between stages.

Summary updated

Abstract

The risk of soft errors due to radiation continues to be a significant challenge for engineers trying to build systems that can handle harsh environments. Building systems that are Radiation Hardened by Design (RHBD) is the preferred approach, but existing techniques are expensive in terms of performance, power, and/or area. This paper introduces a novel soft-error resilient asynchronous bundled-data design template, SERAD, which uses a combination of temporal and spatial redundancy to mitigate Single Event Transients (SETs) and upsets (SEUs). SERAD uses Error Detecting Logic (EDL) to detect SETs at the inputs of sequential elements and correct them via re-sampling. Because SERAD only pays the delay penalty in the presence of an SET, which rarely occurs, its average performance is comparable to the baseline synchronous design. We tested the SERAD design using a combination of Spice and Verilog simulations and evaluated its impact on area, frequency, and power on an open-core MIPS-like processor using a NCSU 45nm cell library. Our post-synthesis results show that the SERAD design consumes less than half of the area of the Triple Modular Redundancy (TMR), exhibits significantly less performance degradation than Glitch Filtering (GF), and consumes no more total power than the baseline unhardened design.
BibTeX
@article{aketi2020serad,
  author = {Sai Aparna Aketi and Smriti Gupta and Huimei Cheng and Joycee Mekie and Peter A. Beerel},
  title = {{SERAD: Soft Error Resilient Asynchronous Design Using a Bundled Data Protocol}},
  journal = {IEEE Trans. Circuits Syst. I Fundam. Theory Appl.},
  year = {2020},
  url = {https://openreview.net/forum?id=ZxoCiv_vTM},
}

Loading related content...

Figure