Nathanael Mohanu

Master of Science in Computer & Embedded Systems Engineering

Bachelor of Science in Computer Science · GPA 7.9/10

Bachelor thesis · Leiden Institute of Advanced Computer Science (LIACS), Leiden University · Grade 8/10

• Captured 160,000 power traces from four crypto co-processors (two AES, Ascon, Xoodyak) on an FPGA prototype (CW305, ChipWhisperer Husky) with interleaved fixed-vs-random acquisition
• Wrote custom C firmware for the Ibex controller core and Python pipelines for trace capture and for TVLA, SNR and CPA analysis
• Showed significant leakage in all four co-processors (|t| up to 193 vs. a 4.5 threshold) and recovered the full AES-128 key from both AES cores via first-order CPA (roughly 2K–3K traces)

Rust · no_std · STM32 · UART · Python

• Wrote a slotted bootloader (under 16 KB) in no_std Rust for an STM32F411 (Cortex-M4) that validates an application's vector table, sets VTOR, and jumps to it
• Designed a CRC-32 framed UART update protocol with a byte-at-a-time parser and offset-based retries; an update is accepted only if the CRC of the image stored in flash matches
• Built a flash driver with sector guards that refuse writes to bootloader sectors, plus host-side unit tests for the CRC, parser and sector map

C++ · BMI2 · AVX-512 · PGO

• Built a specialized single-threaded perft engine tuned purely for throughput: 2.4 billion positions/s on perft(7) without hashing (1.3 s on a Ryzen 9 7900X), nearly 5x faster than Sunsnake
• Removed runtime branching via templates specialized on side to move, depth and en passant, PEXT-indexed attack tables, precomputed pin/check masks, and AVX-512 setwise counting of slider and knight moves
• Verified exact perft counts through depth 9 (2.4 trillion positions) and the perftcheck corpus (561,641 cases at depths 1–4, 140,410 at depth 5); tests run in CI

C++20 · CMake · BMI2 · PGO

• Designed a C++20 CMake library with the core of a chess engine: legal move generation, make/unmake and incremental Zobrist hashing, so a search algorithm can build directly on it
• Reached about 500 million positions/s on perft(7), a benchmark counting all legal move sequences to a fixed depth (3.2 billion in 6.4 s, Ryzen 9 7900X), using BMI2 PEXT/PDEP attack lookups and PGO, single-threaded
• Verified exact perft counts through depth 8 and fuzz-tested against Stockfish on randomized games; perft checks and self-tests run in CI on every push

Lyceo · part-time · Netherlands

• Teach exam preparation in mathematics and English and supervise homework for secondary-school students