Hardware Podcasts

Chip Design for Fun (and Waffles) with Julia Desmazes

The Amp Hour · 9 Apr 2026 · 59m · Episode 721

What this episode covers

Julia Desmazes, a CPU and RTL designer, discusses her experience building custom ASICs for fun through the Tiny Tape Out program, including a recent 'two weeks until tape out' challenge. She explains the open-source silicon stack, detailing how tools like OpenROAD, PDKs, and Wafer Space have lowered the barrier to entry for individuals to tape out chips. Desmazes covers the technical challenges of designing AI accelerators, specifically systolic arrays, and the critical importance of data movement and floating-point optimization in hardware design.

Most relevant to

makers and hobbyists

Topics

Mentioned

Companies
GlobalFoundries, ARM
Tools
OpenROAD, OpenOCD, SEGGER, LiberLane
Also
Tiny Tape Out, Wafer Space, Skywater 130nm, RISC-V

Related episodes from other shows

Other shows on the same ground, closest first.

  • Embedded.fm · 1h 15m

    454: Printf Hello

    Uri Shaked, creator of Wokwi and Tiny Tapeout, explains how Tiny Tapeout democratizes ASIC design by leveraging open-source PDKs and a shared mask strategy to offer chip manufacturing for around $100. He details the technical challenges of building Wokwi, a web-based simulator for embedded systems, including reverse-engineering processor instruction sets and complex peripheral behaviors like SHA and SPI. Shaked also outlines his business strategy, which involves selling space on silicon via Tiny Tapeout and offering Wokwi through a freemium model, VS Code integration, and an on-premise CI/CD solution for professional users.

    Embedded firmwareChips, FPGAs & semiconductors
    8 chapters

  • Embedded.fm · 48m

    105: Some Paths Are More Suicidal

    Manny Wright, Senior FAE Manager at Cortus, explains the process of designing custom processor IP, from RTL to silicon fabrication. He details the high costs and long timelines (1.5+ years, millions of dollars) involved in tape-out, the role of FPGAs like the Xilinx Spartan in early development, and the importance of configurable cores for power and cost efficiency. Wright also discusses security features, debugging methods like JTAG and trace, and the challenges of protecting intellectual property from reverse engineering.

    Chips, FPGAs & semiconductorsMicrocontrollers & dev platforms
    6 chapters

  • Moore's Lobby · 1h 01m

    Synopsys' Stelios Diamantidis Is Bringing the Magic of AI To Chip Design

    Stelios Diamantidis, Distinguished Architect at Synopsys, discusses the evolution of chip design from manual processes to AI-driven optimization. He recounts his career journey, including founding two EDA startups focused on verification and Silicon Lifecycle Management (SLM), before joining Synopsys to lead the development of DSO.ai. Diamantidis explains how reinforcement learning is used to solve intractable physical design problems, optimizing power, performance, and area (PPA) for diverse verticals like CPUs, GPUs, and automotive chips.

    AI & machine learningEDA & CAD tools
    8 chapters

  • The Circuit · 42m

    Episode 67: A Conversation with Aart de Geus on EDA and the Future of Silicon Design

    Aart de Geus, co-founder and executive chair of Synopsys, discusses the evolution of electronic design automation (EDA) and the future of silicon design. He explains how the industry shifted from analog to digital design, enabling the automation that drove Moore's Law, and details the transition to the Angstrom era with technologies like FinFETs and backside power delivery. De Geus also highlights the rise of systemic complexity through chiplets and vertical stacking, and the critical role of AI in optimizing these complex multi-die architectures.

    EDA & CAD toolsAI & machine learning
    8 chapters

  • Oxide and Friends · 1h 31m

    AI Disruption: DeepSeek and Cerebras

    Andy Hock and James Wang of Cerebras discuss how their wafer-scale silicon executes open-weight models like DeepSeek R1 Distill Llama 70B faster than any number of GPUs, achieving 57x speedup by keeping all 44GB of model weights in on-chip SRAM. They explain that the release of efficient, open models like DeepSeek and Llama transformed Cerebras from an exotic supercomputer vendor into a provider of commodity inference APIs, standardizing the unit of sale as tokens. The guests highlight that hardware-software codesign and the shift toward reasoning models allow smaller teams to compete with well-funded incumbents by optimizing algorithms for specific architectures rather than relying on brute-force scaling.

    AI & machine learningChips, FPGAs & semiconductors
    10 chapters

The summary, chapters and topics are generated from the episode’s audio by this directory, not written by The Amp Hour. The episode itself belongs to its publisher.