Sharath Pendyala

Sharath Pendyala

FPGA and hardware security engineer · PhD candidate at NC State

I design hardware for post-quantum cryptography. My PhD research at NC State focuses on fast, secure FPGA and ASIC implementations of FALCON (FN-DSA) and FrodoKEM, and on using large language models to help generate cryptographic hardware. Before that, I spent almost a decade in industry building FPGA designs for satellite base stations, 5G and Wi-Fi test equipment, and high-speed Ethernet.

Select a module to jump to the paper or role behind it.

Research

I’m a third-year PhD candidate in computer engineering at North Carolina State University, working with Aydin Aysu as a research assistant since August 2024. My work sits where post-quantum cryptography meets hardware. I build fast, constant-time FPGA implementations of FALCON (FN-DSA), the NIST-selected lattice signature scheme, study how large language models can generate and optimize cryptographic hardware, and create free, hands-on teaching material for hardware security. I release my designs as open source wherever I can.

  1. FPT 2026Honolulu, Hawaii, USAIEEEAccepted

    FalconFloat: An Open-Source, Specialized Floating-Point Unit for FALCON Signature Generation on FPGA

    Sharath Pendyala, Rahul Magesh, Elif Bilge Kavun, Aydin Aysu

    The first open-source floating-point unit built specifically for FALCON, written in plain Verilog. It drops the IEEE-754 features FALCON never uses (NaN, infinities, subnormals, and exceptions), adds a dedicated floor path, sets latency through multicycle-path constraints instead of pipeline registers, and shares a configurable 64-bit Comba or Karatsuba mantissa multiplier. Placed inside the fastest existing SamplerZ designs on a Zynq UltraScale+, it speeds up sampling by 24–35% and cuts total signature generation time by 15–21%.

  2. GLSVLSI 2026Canandaigua, NY, USAACM

    Microelectronics Systems Education – CHASE: A Cloud-Native Platform for Hardware Security Education

    Rahul Magesh†, Amisha Srivastava†, Sharath Pendyala, Samit Miftah, Aydin Aysu, Kanad Basu

    Free hardware security labs that run in Google Colab with open-source tools, replacing oscilloscopes, fault injectors, and FPGA boards with simulation that keeps the cause-and-effect link between hardware behavior and security. Six modules take students through recovering AES-128 keys with correlation power analysis and differential fault analysis, characterizing Arbiter PUFs, building hardware Trojans, detecting anomalies with hardware performance counters, and breaking logic locking with SAT attacks, at more than 98% lower cost than a traditional lab.

  3. VLSI-SoC 2025Puerto Varas, ChileIEEE

    Deus Ex LLMs: AI vs Humans in Post-Quantum Cryptographic Hardware Code Generation

    Ethan Cornett†, Rahul Magesh†, Sharath Pendyala, Aydin Aysu

    A first look at whether LLMs can write HLS for FALCON’s SamplerZ, comparing ChatGPT, Claude, DeepSeek, Gemini, and Grok against hand-written RTL. The best design came within 4% of the fastest hand-designed sampler’s execution time, and one model found a new optimization: replacing FALCON’s integer exponential approximation with a fixed-latency floating-point exponential. The paper also documents where the models went wrong, from single-precision defaults to overfitting their own testbenches.

  4. ePrint 2025IACR

    Accelerating FALCON: Speed Records for FALCON’s SamplerZ on Xilinx FPGAs

    Sharath Pendyala, Rahul Magesh, Elif Bilge Kavun, Aydin Aysu

    A high-speed design of SamplerZ, the discrete Gaussian sampler that dominates FALCON signing time. It introduces Estrin’s scheme for the exponential approximation, a constant-latency BerExp built on a floating-point exponential core, a faster flooring circuit, and a study of how many parallel sampling loops are worth their area, measured with a new rejection-adjusted execution time metric. Against the previous fastest FPGA design, sampling time drops 55–81% and estimated signature generation time 36–53%, and the fastest version beats a state-of-the-art 28 nm ASIC sampler on total sampling time.

  5. LightSec 2025İstanbul, TürkiyeSpringer

    A Comparison of Unified Multiplier Designs for the Falcon Post-Quantum Digital Signature: A Case Study of Baseline, Tiling, Comba and Karatsuba Approaches for FPGA and ASIC

    Rahul Magesh, Modini Ayyagari, Sharath Pendyala, Aydin Aysu

    FALCON needs one 64-bit multiplier that can serve both floating-point and integer work at one multiplication per cycle. This paper builds it four ways (schoolbook baseline, asymmetric tiling, Comba, and Karatsuba) on a Virtex-7 FPGA and a SkyWater 130 nm ASIC. Karatsuba is the most area-efficient on both platforms, tiling uses about 36% less energy per multiplication on the FPGA, and Comba draws 22.8% less power than the baseline on the ASIC.

  6. Under review

    Refining LLM Hardware Generation: A Two-Level Prompting Approach for Post-Quantum Cryptographic Coprocessors

    Sharath Pendyala, Rahul Magesh, Daniel Rodriguez Arantegui, Sai Vittal Jadav, Elif Bilge Kavun, Aydin Aysu

    An automated flow in which ChatGPT, Claude, Gemini, and Grok generate synthesizable HLS and Verilog RTL for SamplerZ with no human intervention. An algorithm-specific prompt drives generation, and an algorithm-agnostic repair prompt, distilled from more than 80 failed builds, fixes precision, overflow, and constant-time mistakes, while independent known-answer testbenches keep the models from overfitting. The RTL flow cuts development time by more than 40× compared with expert hand design, the generated designs use over 40% less area than prior hardware-only implementations, and the models surfaced two optimizations worth carrying into hand-written RTL.

† Equal contribution

Experience

I’ve spent almost a decade turning specifications into working FPGA hardware: satellite base-station modems at Hughes Network Systems, 5G NR and Wi-Fi 6/7 test equipment at Keysight Technologies, and, as an independent contractor, ophthalmic imaging electronics and Keysight’s M8000 bit-error-rate tester. Most of that work sat where high-speed datapaths, timing closure, and lab bring-up meet, and it shapes how I approach cryptographic hardware in my PhD today.

Aug 2024 – present

Research Assistant

North Carolina State University, Raleigh, NC

Post-quantum cryptography hardware, hardware security, and LLM-aided hardware design, with a focus on FALCON (FN-DSA) and FrodoKEM.

  • Teaching and lab assistant for the undergraduate course Design of Complex Digital Systems on FPGA and the graduate course Cryptographic Engineering and Hardware Security.
Oct 2023 – Jun 2024

Independent FPGA Contractor

Medical imaging and Keysight Technologies

  • RTL for scanning control and safety electronics on a Zynq UltraScale+ (AMD K26 SOM) for scanning laser ophthalmoscopy and optical coherence tomography.
  • Keysight M8000 BERT
    • Implemented partial reconfiguration for fast feature switching, with floorplanning and timing closure for each reconfigurable feature.
    • Designed a cascaded PLL structure for precise clock generation with limited resources, and parts of the IEEE 802.3 100G/200G/400G Ethernet FEC at the physical layer.
Nov 2018 – Jun 2023

R&D IC Engineer

Keysight Technologies, Morrisville, NC

Baseband RTL for the 5G NR and IEEE 802.11ax/be physical layers on Xilinx FPGAs, from simulation through lab and customer-site testing.

  • Designed the resource-mapping module supporting SRS 8×8 MIMO on a Virtex UltraScale+ for the 5G NR UE emulator.
  • Brought up 5G NR millimeter-wave over-the-air test setups in Keysight’s lab and at customer sites including Verizon, and measured FR2 throughput against a UXA signal analyzer at distances up to 16 feet.
  • Cut Wi-Fi receiver end-to-end latency to meet the 802.11ax SIFS requirement, including clock-domain-crossing and FFT changes; added 160 MHz bandwidth support and 802.11be EHT PPDU decoding.
  • Ported the 802.11ax MAC and PHY from the Wi-Fi test platform (AXE) to the 5G test platform (UXM) as a proof of concept, reaching MAC and IQ-sample loopback.
  • Built multi-FPGA, multi-PPDU simulation testbenches and analysis tools for debugging the receiver.
Jul 2015 – Nov 2018

Hardware Engineer 3

Hughes Network Systems, Germantown, MD

Architected the baseband modem RTL for a satellite 3G base station and designed the next-generation satellite base-station subsystem on a Virtex-7.

  • Created a DDR memory organization and arbitration scheme around a modified Xilinx MIG that averaged over 70% efficiency, with peaks of 93%, in a live system, including offset-address bursts to absorb Doppler-induced sample offsets.
  • Added built-in self-test, memory rotation to limit wear under continuous use, and design-for-test access through the DDR arbiter.
  • Packed multiple asynchronous, width-converting FIFOs into a single block RAM to improve timing closure and area.
  • Designed digital interpolators and FIR filter paths, wrote Vivado TCL and Bash scripts that raised build success rates, and led integration, rack design, and agency testing.

Skills

Design and research
FPGA architecture, RTL design, hardware security, post-quantum cryptography, algorithm development, timing and area optimization, digital signal processing
Languages
Verilog, SystemVerilog, VHDL, HLS, Python, TCL, C
Design tools
Vivado, QuestaSim, ModelSim, MATLAB
Test and measurement
Signal Studio, WaveJudge, vector signal analyzers, signal generators, Wireshark
Version control
Git, ClearCase

Education

  • PhD, Computer Engineering (Hardware Security)

    North Carolina State University · 2024 – 2028 (expected)

  • MS, Computer Engineering

    North Carolina State University · 2015 · GPA 3.83/4

  • BTech, Electronics and Communication Engineering

    National Institute of Technology, Tiruchirappalli · 2013 · GPA 8/10

Contact

I’m happy to talk about post-quantum hardware, FPGA design, research collaborations, or roles in hardware security.

University
spendya@ncsu.edu
Personal
sharathpawan@gmail.com
LinkedIn
linkedin.com/in/sharathpawanp
GitHub
github.com/sharathpend