TL;DR
Researchers have identified a critical "weak link" in radiation-hardened DICE flip-flops: the auxiliary set-control circuitry. By implementing a spatially redundant set-signal path, they successfully eliminated Single Event Upsets (SEUs) in a 40nm process, achieving zero errors under heavy-ion simulation with a manageable ~14% overhead in area and power.
What Problem Does This Paper Address?
In the harsh environment of space, high-energy particles (cosmic rays) can strike semiconductor devices, causing Single Event Effects (SEE). A common issue is the Single Event Upset (SEU), where a bit of data in a flip-flop flips from 0 to 1 or vice versa. To combat this, engineers use Radiation Hardening By Design (RHBD), specifically the Dual Interlocked Storage Cell (DICE) architecture, which uses redundancy to self-correct errors.
However, most research focuses on the storage core itself. This paper highlights a neglected vulnerability: the "Set" circuit. Even if the storage core is hardened, a particle strike on the control path that tells the flip-flop to "set" to high can generate a Single Event Transient (SET)—a momentary false pulse. This pulse can "trick" the hardened core into changing state, effectively bypassing its built-in protections.
Methods
The authors employed a sophisticated multi-level simulation framework to analyze and fix this vulnerability in a SMIC 40nm process:
- TREES Simulation: Used for full-layout heavy-ion strike scanning. It maps the physical layout to identify "hotspots" where a particle strike is most likely to cause a failure.
- TCAD-Calibrated SPICE: The researchers used Technology Computer-Aided Design (TCAD) to model the physical physics of a particle strike on a transistor. They then used these precise current models in SPICE (circuit simulation) to see how the resulting voltage pulses propagate through the logic.
- Redundant Design: They proposed a "set-redundancy" scheme. Instead of one set signal, they created two separate paths (S1 and S2) that are physically separated on the chip layout. Because they are far apart, a single particle strike is unlikely to affect both simultaneously.
Datasets
As this is a hardware design and simulation paper, it does not use traditional machine learning datasets. Instead, the "data" consists of:
- Layout Strike Locations: 1,691 discrete strike points for the baseline design and 1,919 points for the hardened design.
- Simulation Parameters: Heavy-ion strikes with a Linear Energy Transfer (LET) of 37 MeV·cm²/mg, representing a typical harsh space environment.
- Process Node: SMIC 40 nm CMOS technology models.
Key Insights and Ideas
- The "System Shortcut": The most significant insight is that the control logic (Set/Reset) can act as a "shortcut" for radiation-induced errors. Hardening the storage core is useless if the "Set" button can be pushed by a stray ion.
- Spatial Isolation is Key: The researchers found that simply adding a second signal isn't enough; the two paths must be physically separated in the layout (spatial redundancy) to prevent "charge sharing," where a single strike spreads its energy across both paths.
- Leveraging DICE Self-Recovery: By splitting the set signal, a strike on one path only affects one node of the DICE core. Because the DICE architecture is designed to recover from single-node upsets, it automatically ignores the false "set" pulse.
Key Results
The proposed RHBD strategy showed dramatic improvements in reliability:
- Error Elimination: In the baseline design, 18 specific strike locations caused an SEU. In the hardened redundant design, zero SEUs were observed across all 1,919 strike locations.
- SEU Cross-Section: The sensitive area (cross-section) dropped from $1.7 \times 10^{-9} \text{ cm}^2$ to zero.
- Manageable Overhead: The hardening came at a cost of a 14.3% increase in area and a 14.5% increase in power consumption.
- Timing Stability: Critical timing parameters (like signal delay) remained largely unaffected, making it a practical drop-in replacement for standard cell libraries.
Limitations and Future Work
While the simulations are promising, the authors note a few areas for further exploration:
- PVT Variations: The current study focused on specific conditions. Future work needs to account for variations in Process, Voltage, and Temperature (PVT) to ensure the design is robust across all operating envelopes.
- Monte Carlo Analysis: To further refine the reliability statistics, more extensive Monte Carlo simulations (randomized strike variables) are suggested.
- Physical Validation: The ultimate test will be fabricating the design on a test chip and subjecting it to actual heavy-ion beam testing in a laboratory.
Summary updated