AMD Announces Versal Tamper-Resistant Adaptive SoCs With a Backside Protection Die and ML-DSA and SLH-DSA Accelerators on 12 November 2026, Sampling in 2027: Readiness Not Applicable
Defense Henry Quentir Defense Henry Quentir

AMD Announces Versal Tamper-Resistant Adaptive SoCs With a Backside Protection Die and ML-DSA and SLH-DSA Accelerators on 12 November 2026, Sampling in 2027: Readiness Not Applicable

On 12 November 2026 AMD announced the 2VE3358TR, 2VE3558TR and 2VE3858TR, Versal AI Edge Series Gen 2 adaptive SoCs with a sensor-based backside protection die bonded to the compute die by 3D active-on-active technology, so the device refuses to boot if the die is tampered with or removed. The same parts add ML-DSA and SLH-DSA hardware accelerators, a hybrid boot flow, an improved key vault, a memory element aging sensor and a second glitch detector, with samples in the second half of 2027 and production in the first half of 2028. The post sets the announcement beside AMD security bulletin AMD-SB-8034 of 2 September 2026 and the Bochum Leakback CRC paper, which showed optical backside probing reading encrypted bitstreams on 7-Series FPGAs, reads the development as a defensive post-quantum root of trust in hardware whose readiness cannot yet be placed on the ladder, and lists the silicon, threat model, independent evaluation and CNSA 2.0 calendar a program office should wait for.

Read More
Georgia Tech's Gupta Puts a Ceiling on What One Twiddle Fault Leaks From ML-KEM and ML-DSA Key Generation, arXiv 2610.30383, 21 October 2026: 254 of 256 Coefficients Exposed on Paper, TRL 2
Defense Henry Quentir Defense Henry Quentir

Georgia Tech's Gupta Puts a Ceiling on What One Twiddle Fault Leaks From ML-KEM and ML-DSA Key Generation, arXiv 2610.30383, 21 October 2026: 254 of 256 Coefficients Exposed on Paper, TRL 2

Georgia Tech's Chakshu Gupta proves, in Lean 4, how much a single corrupted twiddle constant in the number-theoretic transform exposes during ML-KEM and ML-DSA key generation: a rank that halves by layer from 128 to 2, a ceiling of 254 of 256 coefficients with one fault per layer, and two coefficients no static twiddle fault can reach. The Defense Monitor reads what the fault attack bound means for a hardware root of trust that regenerates keys from a stored seed, how it extends the 2023 electromagnetic attack on pqm4, and what a module buyer should ask next.
Read More
Shandong University's Chen and Li Recover 16 Private-Key-Dependent Bits From BIKE's BGF Decoder by Deep-Learning Power Analysis at 95.68 Percent, Cybersecurity 9:215, 16 September 2026
Defense Henry Quentir Defense Henry Quentir

Shandong University's Chen and Li Recover 16 Private-Key-Dependent Bits From BIKE's BGF Decoder by Deep-Learning Power Analysis at 95.68 Percent, Cybersecurity 9:215, 16 September 2026

On September 16, 2026, the journal Cybersecurity published a paper by Geng Chen, Yanbin Li and colleagues at Shandong University that reads a private-key-dependent 16-bit sequence out of the BIKE post-quantum decoder from a single power trace on an ARM Cortex-M4 board, at 95.68 percent perfect recovery. We read the deep-learning side-channel attack as an offensive extraction method, weigh what it does and does not transfer to the ML-KEM and HQC code a force will field, and set out the missing key-recovery stage, the untested countermeasures and the single-board evidence that stand between this post-quantum implementation security result and anything a program office would plan against.

Read More
The Post-Quantum Migration Reaches the Silicon Layer
Defense Henry Quentir Defense Henry Quentir

The Post-Quantum Migration Reaches the Silicon Layer

In one August week, a Wuhan university team released a 28-nanometer post-quantum cryptography chip with a naval research institute in the room, and a Canada-Taiwan consortium validated a compute-in-memory rival for the same NIST algorithms. A close look at what a hardware root of trust means for forces whose platforms outlive their ciphers, and at the evaluation gates still standing between both chips and a program office.

Read More