Hardware Podcasts

AI PCB Routing Without the Epic Fails: Inside DeepPCB

OnTrack · 16 Sept 2026 · 38m · Episode 354

AI in practice: 5 real uses of AI described

What this episode covers

Alain-Sam Cohen and Nabil Chouba of InstaDeep’s DeepPCB explain how their reinforcement-learning routing engine, paired with the natural-language orchestrator Cooper, detects design intent like net classes and impedance before routing. They demonstrate how the tool identifies missing context, such as unspecified reference planes and USB impedance mismatches, allowing the AI to flag and fix errors that classic autorouters miss. The guests argue that task-based AI, which focuses on specific optimization problems rather than generating entire designs from vague prompts, is the viable path for hardware automation. They also discuss how DeepPCB integrates with code-based PCB tools by acting as the placement and routing backend for schematic generation.

AI in practice

Where this conversation describes AI being used to build or deliver a product. Part of the AI in practice tracker.

DesignDeepPCB (Cooper)

An LLM-based orchestrator named Cooper analyzes an underspecified PCB design file to detect missing context like net classes and decoupling capacitors, then automatically applies corrections such as changing layer types to power planes before routing begins.

“Cooper is just the LLM that sits on top of the system, right? Cooper is not an autoroter. It's an orchestrator of the DPCB routing engine. So now it's looking at your PCV and this unlocks a bunch of things because so many of our users, again, were frustrated by not being able to have the right configuration.”

Alain-Sam Cohen · Hear it at 26:50

DesignDeepPCB (Cooper)

The AI system detects signal integrity issues, such as incorrect trace width for USB impedance matching based on the board stack-up, and flags these errors to the user, offering to fix them automatically rather than failing silently.

“I check. I tell the, yeah, can you check the signal and take it? Tell me, you have a speed signal and it's differential here. And based on the stack up of the bird, the gap is incorrect, this is not good. It's USB, it needs to be ATR, man, it's not ATR.”

Nabil Chouba · Hear it at 22:42

DesignDeepPCB (Cooper)

The AI detects critical signals to be routed first, identifies decoupling capacitors, and reads schematics to guess design intent, allowing it to configure the autorouter correctly without requiring the user to manually spend hours on configuration.

“we we find those mistakes, we fix them for you, but also we detect the critical signal that they're going to be routed first. We detect the decapling capacitors. We try to detect the pinsica thing. We read the schematics and we try to do some guesses from there, from there.”

Nabil Chouba · Hear it at 22:42

DesignDeepPCB (Cooper)

The system uses an LLM with specific 'skills' to execute tasks like changing layer types, adjusting track widths, and initiating the routing engine for a specified duration, allowing the user to accept or reject suggestions.

“Cooper has skills. One of the skills of Cooper is change, uh, a layer type or change the, A track width, but also put this net in this net test and route your board for 5 minutes or just route this net for 10 minutes. So it's kind of like the analogy you can make.”

Alain-Sam Cohen · Hear it at 26:50

DesignDeepPCB (Routing Engine)

A reinforcement learning-based routing engine iteratively generates and improves PCB trace layouts, converging on a solution that satisfies design constraints and DRC rules without manual intervention after initial configuration.

“It's a sequential decision making problem. Reinforcement learning, as you know, it's trial and error … if you start connecting points where they have to actually go, it's good, if you go into a world that's not that good, and if you end up connecting, but you do 3 turns, uh, violate some cream through clearance rules…”

Alain-Sam Cohen · Hear it at 5:07

Most relevant to

early-stage companies

Topics

Mentioned

Companies
InstaDeep, ST Microelectronics, NXP
Products
DeepPCB, Cooper

Related episodes from other shows

Other shows on the same ground, closest first.

  • Moore's Lobby · 56m

    Solving the Hard Problems: Combining AI and Physics for PCB Design

    Sergiy Nesterenko, founder and CEO of Quilter, explains how his company uses reinforcement learning and computational physics to automate PCB layout, aiming to create a 'compiler' for electronics that handles the entire design scope. He recounts his early career at SpaceX, including an internship watching paint dry and a subsequent board fire that highlighted the reliance on tribal knowledge and the failure of traditional auto-routers. Nesterenko details Quilter's workflow, which integrates with existing CAD tools like Altium to generate layouts for test boards and complex single-board computers, while acknowledging current limitations with high-voltage or ultra-dense server designs. The discussion covers the technical strategy of using neural networks to interpolate simulation results, the company's focus on iterative complexity scaling, and a 10-year vision for enabling continuous integration in hardware design.

    PCB design & layoutAI & machine learning
    9 chapters

  • The Amp Hour · 1h 07m

    Intelligent Routing with Sergiy Nesterenko

    Sergiy Nesterenko, CEO of Quilter.ai, explains how his company uses generative AI and parallel cloud compute to automate PCB layout from schematics. He discusses the company's pivot from hardware sensors to software, drawing on his experience at SpaceX where manual routing became a critical path bottleneck. Nesterenko details how Quilter handles constraints, validates designs using built-in Maxwell equation solvers, and targets specific fabs like JLCPCB and Oshpark to ensure manufacturability.

    EDA & CAD toolsAI & machine learning
    7 chapters

  • The Amp Hour · 1h 00m

    Layout Review with Zachariah Peterson

    Zachariah Peterson, founder of Northwest Engineering Solutions and Altium's technical consultant, discusses the state of PCB design, the industry's shift toward EMC expertise, and the impact of AI tooling. He details Altium's acquisition by Renesas, the critical need for engineers to understand manufacturing constraints and stack-ups, and the risks of relying on semiconductor data sheets for EMC guidance. Peterson also shares insights from his 'Layout Review' content series, highlighting common DRC errors and the value of mentorship in a lean industry.

    EDA & CAD toolsSensors & measurement in products
    9 chapters

  • IoT For All Podcast · 21m

    Can AI Design IoT Hardware? | Flux's Matthias Wagner | Internet of Things Podcast

    Matthias Wagner, Founder and CEO of Flux, explains how his company uses agentic AI to democratize PCB design, allowing non-engineers to create hardware via natural language prompts. He discusses the historical bottlenecks of hardware design, the necessity of iterative back-and-forth for complex devices like smartphones, and the economic advantages of designing custom boards for products like vending machines. Wagner also outlines future integration challenges, including firmware generation and enclosure design, and predicts that lowering the cost of iteration will unlock significant innovation in robotics and IoT.

    AI & machine learningPCB design & layout
    8 chapters

  • TRADEOFFS · 1h 01m

    T-25-010 : Ioannis Papamanoglou and Narayan Powderly - Engineers transforming how we design, test, and build hardware with atopile

    Narayan Powderly and Ioannis Papamanoglou, co-CEOs of atopile, discuss their command-line circuit design tool that brings software development practices to hardware engineering. They explain how their frustration with slow board development cycles at Tesla and smart bike lock projects led them to create a domain-specific language that captures design intent rather than just schematics. The guests detail how atopile integrates with KiCad for layout, leverages LLMs for rapid iteration, and aims to empower full-stack engineers by automating the drudgery of component selection and validation.

    EDA & CAD toolsAI & machine learning
    9 chapters

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