The Amp Hour

Eric Pan is the founder and CEO of Seeed Studio, an open-source hardware and manufacturing platform based in Shenzhen, China. He joins Chris to discuss Seeed’s journey from early open-source roots to modern robotics, navigating the hardware ecosystem in Shenzhen, bridging the gap between semiconductor companies and developers, and how physical AI and low-cost hardware platforms are transforming innovation. Timeline Chris welcomes Eric Pan, founder and CEO of Seeed Studio, to discuss open-source hardware, the evolution of Shenzhen, and the emerging robotics ecosystem. (00:00:15) Eric recounts founding Seeed Studio in Shenzhen in 2008, starting with just two people committed to open hardware. (00:02:45) Eric discusses his background working at Intel and how Seeed’s earliest projects supported artists and interactive installations needing custom electronics. (00:05:30) Chris and Eric reminisce about the landmark 2015 Maker Faire Shenzhen and look forward to its return this December. (00:09:15) How hardware startups, suppliers, and maker spaces collaborate, share resources, and exchange talent across Shenzhen. (00:13:00) How the process of procuring components, tooling, and materials in Shenzhen has transformed over the past decade. (00:16:45) How Shenzhen’s modular electronics supply chain is pivoting to support rapid prototyping and production of robotic platforms. (00:20:30) Looking at Espressif Systems as a standout model of a domestic chip company that built a global market by championing developer communities and open-source software. (00:24:15) Why traditional semiconductor manufacturers are often hesitant to fund grassroots developer support, and the cost of neglecting community adoption. (00:27:45) How silicon vendors like Rockchip engage with open hardware ecosystems and community developers. (00:31:15) Brief audio dropout during the recording session. (00:34:45) The engineering philosophy behind creating the thumb-sized Seeed Studio XIAO microcontroller board series and packing high compute into minimal footprints. (00:35:15) Assessing the rise of “vibe coding” and how generative AI code assistants are influencing firmware development and electronics prototyping. (00:38:00) Exploring whether large language models and software automation threaten hardware platforms or act as powerful accelerators for engineers. (00:40:45) Brief recording interruption. (00:43:00) How rapid tooling and AI-assisted workflows speed up iteration cycles for physical electronics. (00:43:15) How Seeed is moving beyond board manufacturing to act as an ecosystem advisor connecting silicon vendors, factories, and developers. (00:45:15) Eric shares his perspective on viewing Shenzhen’s manufacturing hub as an open supermarket of modular resources rather than a cutthroat arena. (00:47:30) Exploring the shift from pure software models to embodied intelligence, connecting back to Chris’s conversation on Episode 732: Hands-on Physical AI with Kevin Cloutier. (00:49:45) Chris discusses purchasing the SO-101 Robot Arm Kit, which he previously mentioned in Episode 734, and how Seeed stepped in to support DIY builders who kept asking for assembly and calibration help. (00:52:00) How the open-source LeRobot framework is democratizing imitation learning, teleoperation, and robotics training. (00:54:15) Discussing the open-source Pollen Robotics Microduck bipedal robot and reinforcement learning sim-to-real pipelines. (00:56:30) Why sub-$200 open-source kits and standardized software stacks are lowering barriers to robotics education. (00:58:30) Seeed’s collaboration with NVIDIA around Jetson hardware for edge AI and robot control. (01:00:15) How Seeed serves as a vital bridge between chipmakers and global developer communities. (01:01:45) The practical realities of modern factory automation on the ground in Shenzhen. (01:03:00) Exploring the full spectrum of industrial automation—from simple fixtures and cobots all the way up to humanoid robots. (01:04:15) How the Seeed Fusion PCBA turnkey service helps hardware innovators prototype and scale production. (01:05:30)

Kjetil Meisal, CEO and co-founder of Norwegian semiconductor startup ONiO, joins Chris to discuss the development of ONiO.zero—a revolutionary batteryless microcontroller powered entirely by ambient energy harvesting. They dive into Norway’s rich semiconductor heritage, custom RISC-V architecture, sub-microwatt cold starts, paper-based biofuel cells with BeFC, and making “stick and forget” IoT sensors a reality. Timeline Welcome Kjetil Meisal: Chris welcomes Kjetil Meisal, CEO of ONiO, to discuss their vision for truly batteryless microchips. (00:00:15) Stick and Forget: Exploring the core dream of deployment-friendly IoT sensors that run indefinitely on harvested energy without maintenance. (00:02:15) Nordic Silicon Heritage: Kjetil highlights Norway’s deep semiconductor roots and ecosystem spanning Oslo and Trondheim. (00:05:00) Norwegian Chip Companies: A breakdown of global heavyweights built on Norwegian acquisitions, including Texas Instruments (former Chipcon), Silicon Labs (former Energy Micro), ARM Trondheim (former Faltung), Microchip (Atmel AVR team, discussed on The Amp Hour #633), Nordic Semiconductor, Novelda, and startups like ONiO. (00:07:45) The ONiO.zero Platform: Banner specs of the ONiO.zero MCU—a custom RISC-V CPU core, multi-protocol 2.4 GHz radio, power management unit, and multi-domain energy harvesting. (00:13:00) Energy Storage Options: Moving from ceramic capacitors and supercapacitors to rechargeable cells under a “batteries not required” philosophy. (00:15:45) Going Back to 2016: The origin story of ONiO with co-founders Kjetil Meisal, Runar Finanger, and Vemund Bakken exiting Novelda to solve low-power challenges. (00:19:15) From Fever Patches to Silicon: Transitioning from a single-use temperature monitoring patch concept to general-purpose ultra-low-power silicon. (00:22:15) Mature Process Nodes and In-House IP: Why ONiO chose a robust, mature silicon node and developed 100% of their IP in-house without third-party licensing. (00:25:00) Developer Mental Shift: How clock gating, powered-off blocks, and GCC-based toolchains require firmware developers to think in terms of energy scarcity. (00:27:45) Sub-Microwatt Cold Start Sequence: How the Power Management Unit harvests ambient voltage, triggers ultra-low voltage monitoring loops, and safely boots the RISC-V core. (00:30:00) Pomegranate Seed Power: Demonstrating extreme low-power operation by booting the chip and transmitting Bluetooth signals off citrus fruits, tomatoes, blueberries, and a single pomegranate seed. (00:32:30) Energy Harvesting Sources: Matching ambient indoor light, outdoor solar, Peltier thermal gradients, and RF signals with ultra-efficient silicon. (00:37:00) Managing the Energy Reservoir: Transmitting data in bursts, managing duty cycles, and running a wireless thermostat for two weeks on a supercapacitor in total darkness. (00:40:00) Custom 2.4 GHz Radio Stage: Achieving -100 dBm sensitivity at 3 mA RX and 0 dBm at 6 mA TX with custom RF integration supporting 5+ protocols. (00:42:30) Why RISC-V over ARM?: Why ARM’s licensing fees and instruction restrictions led ONiO to custom RISC-V microarchitecture tweaks, echoing open-architecture discussions on The Amp Hour #637 and The Amp Hour #687. (00:45:30) CoreMark & ULPBench Benchmarks: Breaking down ONiO’s 22 µW/MHz efficiency, 4.1 CoreMark/MHz score, and record 180.75 EEMBC ULPMark score. (00:48:30) Compute Trade-offs: Balancing clock speed vs. leakage and why continuous low-power computing outweighs raw high-performance processing in IoT. (00:52:00) On-Board TinyML and Event Processing: Using lightweight edge signal processing to transmit event triggers rather than continuous raw data streams. (00:54:30) Key Application Domains: Deploying batteryless sensors in precision agriculture, smart building infrastructure, asset tracking, and perpetual biosensing (as previously explored on The Amp Hour #733). (00:58:00) Biofuel Cells with BeFC: Partnering with French startup BeFC to combine paper-based enzymatic sugar fuel cells with ONiO silicon for compostable tracking tags and skin patches. (01:00:15) Customer Engagement & Trade Shows: Partnering with early-access customers, offering QFN packages, and appearing at IFA Berlin, Electronica, GITEX Global, and TechCrunch Disrupt. (01:01:45) Sustainable Electronics: Exploring biodegradable PCB substrates, finding ONiO at onio.com, and following their updates on LinkedIn. (01:02:45)

Dave and Chris catch up on Dave’s visit to the University of Sydney Open Day, the shift toward practical hands-on engineering education, and open-source robotics like the Seeed Studio SO-101 and Hugging Face Micro Duck. They also discuss calculator forensics, Dave mentoring his intern Dylan through voltage regulator selection, AI-assisted PCB layout tools, why Chris wants to be “the Walmart greeter of electronics,” reverse-engineering the Rigol DNA800 network analyzer, Dave’s classic toaster design story, and news from Espressif and Build Australia. Timeline Open Day at Sydney Uni: Dave discusses taking his son Sagan to the University of Sydney Open Day and wondering where all of their engineering graduates end up working. (00:00:45) Where Do Sydney Uni EE Grads Go?: They speculate on why University of Sydney graduates rarely show up in practical engineering job applicant pools, noting high international student enrollment and potential moves into fintech or graduate school. (00:02:30) Practical Engineering Education: Dave expresses surprise at the massive emphasis on practical, first-year hands-on projects across all engineering disciplines at Sydney Uni compared to past decades. (00:05:15) US vs. Australian Engineering Schools: Chris and Dave compare recruitment dynamics, big state universities like Michigan and UIUC, and smaller elite institutions like Rose-Hulman, Olin, and Harvey Mudd. (00:07:30) Larry Sears and Campus Makerspaces: Chris remembers past guest Larry Sears and his donation to build the 50,000 sq. ft. Sears think[box] makerspace at Case Western Reserve, comparing it to institutional makerspaces at Georgia Tech and Duke. (00:10:15) School Calendars & High School STEM: A quick look at Australian vs. US academic calendars and the rise of high school robotics programs like FIRST Robotics and Lego competitions. (00:13:45) Seeed Studio SO-101 Robot Arm: Chris shares his excitement about assembling the open-source SO-101 robot arm kit from Seeed Studio for trade show demos, using leader-follower training and Vision-Language-Action (VLA) models. (00:15:30) Hugging Face & Micro Duck Robot: They talk about Nvidia’s acquisition memes ($12.96B ASCII easter egg decoding to the hugging face emoji) and Pollen Robotics / Hugging Face’s upcoming $400 open-source Micro Duck robot kit. (00:17:45) Calculator Forensics: Dave describes explaining 4-bit processors and datamath.org Calculator Forensics to his son Huxley, testing internal register accuracy using trigonometric function chains like arcsin(arccos(arctan(tan(cos(sin(9)))))). (00:20:15) Mentoring Dylan on Voltage Regulators: Dave talks about guiding his second-year intern, Dylan, through parametric searching to pick a 3V fixed regulator for an ESP32-based product. (00:24:00) LDO Stability & Package Selection: Dave delivers an impromptu masterclass on SOT-23 vs. SO-8 packages, LDO loop stability at microamp quiescent currents, transient responses for LCD/Bluetooth burst currents, and dielectric differences between Y5V and X7R capacitors. (00:27:30) AI PCB Design & Astra Announcement: They review OpenAI’s Astra demo showing automated KiCad schematic generation and live 4-layer PCB layout routing. (00:33:15) The Walmart Greeter of Electronics: Chris explains why he refuses to gatekeep hardware design as AI layout tools lower the barrier to entry, preferring to be “the Walmart greeter of electronics” welcoming newcomers into the field. (00:35:15) Debugging & The Middle Gap: They analyze the gap between simple AI-generated microcontroller boards and complex high-speed multi-gigabit hardware, questioning where new engineers will learn root-cause debugging when boards fail. (00:37:30) History of Auto-Routers: Dave points out that automated PCB placement, auto-routing, and neural net hype have existed in EDA tools for over 30 to 40 years. (00:40:15) KiCad, Bezier Curves, and Trace Cloning: Dylan asks why KiCad’s schematic editor supports Bezier curves, and Chris highlights past guest Ian Scott Johnson contributing trace characteristic cloning features to KiCad. (00:43:00) The Power of Open Formats: Chris emphasizes how text-based open formats like KiCad, FreeCAD, and Blender allow AI tools to hook into hardware design, referencing past guest Robert Fennis and open FEM solvers. (00:44:45) AI in Semiconductor Design: They touch on an Asianometry video covering Synopsys and Cadence AI chip design tools, stock market drops, and token-based pricing models. (00:46:30) Reverse Engineering the Rigol DNA800 Scope: Dave details tearing down and reverse-engineering the RF board on Rigol’s DNA800 (Digital Network Analyzer / VNA scope). (00:48:00) RF PCB Layout Techniques: They examine gold-shielded PCB arenas, surface traces vs. inner-layer vias, and unusual stripline couplers feeding IF signals back into the local oscillator. (00:50:45) Leaving Mistakes in Videos: Dave discusses accidentally misidentifying a DAC as an ADC in his teardown video and leaving the mistake in to show authentic reverse-engineering. (00:54:00) The Toaster Designer Story: Inspired by the book Red Rising calling robots “toasters,” Dave retells his classic story of turning down a job designing a constant-temperature toaster controller because “toaster repairman/designer” was an old forum running joke. (00:55:45) The Toaster Chip Legacy: Chris references Episode 541 where they previously discussed dedicated toaster ASICs. (00:58:00) Edge Lab Live at Electronica: Chris mentions Edge Lab Live running a hands-on engineering lab at Electronica in Munich this November. (00:59:15) Espressif Launches Linux BSP: Chris highlights Espressif quietly launching a Linux BSP for the ESP32-P4 chip, as shared on the Contextual Electronics forum thread. (00:59:45) Build Australia YouTube Creator Job: Dave shares a job opening from Build Australia seeking a YouTube creator to visit and film local Australian hardware and manufacturing companies. (01:00:30)

Chris welcomes Andy Kong to discuss “Chargerless” and how to build ultra-low-power wearable health trackers that never need wall charging. (00:00:15) From Hacker Mindset to Startup Founder: They discuss Andy’s background in biosensing, staying with Samy Kamkar, and bringing a pragmatic “get things done” hacker mindset to founding Chargerless in San Francisco, as documented on Andy’s personal website. (00:02:10) Solving the Wearable Charging Bottleneck: Andy explains how frustration with daily Apple Watch charging led him to design a wearable requiring only one minute of daily sunlight exposure, noting that modern wearables are bottlenecked by battery volume rather than logic chip size. (00:04:30) Deliberately Underpowered MCUs: Chris and Andy explore system architecture trade-offs, focusing on Chargerless’s choice of an underpowered microcontroller paired with lean firmware rather than feature-heavy, power-hungry processors. (00:07:15) Intelligent Sampling vs. Continuous Monitoring: They contrast Whoop and Apple Watch’s continuous PPG power drain with Chargerless’s periodic sampling approach, which reserves high-frequency tracking for active workout modes. (00:09:45) The Limits of Commercial Sensor ICs: Why off-the-shelf health ICs from TI and Analog Devices consumed 100 times too much power due to unoptimized ADCs, inefficient amplifiers, and fixed pulse timings. (00:12:20) EKG vs. Wrist PPG: Explaining how multi-lead chest EKGs measure electrical heart potential vectors, whereas wrist Photoplethysmography (PPG) uses green light reflection to detect oxygenated blood volume changes. (00:15:05) Badge Life and DIY Biometrics: Chris and Andy reminisce about community hardware projects like Chaos Camp’s Card10 EKG badge, Kerry Scharfglass’s IR-synchronized dragonfly badge, and Chris’s earlobe PPG sensor that flashed LEDs inside a Solo cup based “badge”. (00:17:50) Custom Discrete PPG Frontend: How Andy replaced commercial ICs with a discrete circuit that strobes a green LED 50 times per second for microsecond bursts, cycling both the LED and transimpedance amplifier. (00:20:40) Eliminating Decoupling Lag: Stripping decoupling capacitors removed turn-on delays, while using two amplifier stages avoided high-capacitance digital potentiometers. (00:23:10) Pivoting Away from BCI: Andy describes his college research on brain-computer interfaces (BCIs) with OpenBCI and why he abandoned non-invasive EEG due to microvolt-level signal noise from cable movement. (00:25:55) Quantified Self & Personal Informatics: Discussing how self-trackers capture environmental data to identify root causes of complex health symptoms. (00:28:30) Low-Power Bluetooth Data Batching: How Bluetooth Low Energy accounts for less than 5% of overall power draw by batching 10 KB payloads every 10 minutes for fast burst transmissions to the phone. (00:31:15) Form Factors and Ring Safety: Comparing Whoop, Oura, Garmin, and Fitbit Air trackers, while covering smart ring battery constraints, Samsung battery recall testing, and degloving risks. (00:34:00) Whoop’s Slide-On Battery: How Whoop designed a slide-on battery pack to charge while being worn, avoiding the user drop-off caused when devices are taken off to charge. (00:36:45) Solar Harvesting & Laser Dicing: Chargerless uses boost converters to harvest power from amorphous solar cells, which Andy initially custom diced with laser cutters at MIT. (00:39:20) Bare-Metal Registers & 28-Hour Days: Transitioning from dev boards to custom PCBs required low-level MCU register programming, custom Makefiles, and a 28-hour daily cycle sleep study in tinfoiled rooms. (00:42:10) Pricing & Open Data Access: Chargerless is rolling out a pre-production run of a few hundred units for $199 with no subscription fees on getchargerless.com and open data export capabilities. (00:45:00) Power Over Skin (Body Channel Communication): Andy breaks down his Power over Skin research project, which sends 40–100 MHz RF power safely through human skin using the body as an antenna. (00:47:45) Microwatt Energy Harvesting: Utilizing Skyworks low-forward-drop diodes, tank circuits, and rectifiers to harvest microwatts for powering LED earrings, ring joysticks, and skin patches. (00:50:30)

Kevin Cloutier is the North American Lead for Physical AI at CapGemini, a global engineering consulting firm. He is a computer engineer who is passionate about taking complex computations to the edge. He joins Chris to discuss the transition from legacy, statically programmed factory robotics to modern, imitation-learning-based “Physical AI,” building and calibrating the affordable $200 3D-printed SO101 robot arm using the LeRobot open-source framework or Intel’s Physical AI Studio, and orchestrating complex robotic systems entirely on local edge hardware. Timeline Chris welcomes Kevin Cloutier to discuss “Physical AI” and how it relates to robots and things that move in the physical world. (00:00:15) Meeting at Embedded World: They reminisce about meeting at the Canonical booth in front of a robot demonstration, which you can see in the Embedded World Demonstration Video. (00:02:10) Trade Shows and Geography: Chris and Kevin discuss Embedded World in Nuremberg, Germany, upcoming SPS, and the high concentration of manufacturing robots in Europe compared to the US. (00:03:50) The “Unicorn Developer” Shift: Kevin talks about his background in computer engineering and how the “unicorn developer” has shifted from the full-stack web developer of the 1990s to someone who can operate, program, repair, and train physical robots. (00:06:05) Factory Robots vs. Edge Cases: Contrasting legacy, statically programmed factory robots (like ABB or Universal Robots hanging car doors) with modern generative AI approaches capable of handling real-world edge cases. (00:09:15) The Robotics Software Stack: Demystifying the layers above motor drivers, including Real-Time Operating Systems (RTOS), Linux, and message-passing frameworks like ROS (Robot Operating System) and ROS 2. (00:12:40) Orchestrating Systems of Systems: Kevin describes playing tic-tac-toe using Vision Language Action (VLA) models, where a higher-level camera and computer vision system orchestrate the coordinates for the movement model. (00:15:10) The Evolution of Compute: How modern silicon, integrated GPUs, and SOCs have allowed the massive, heavy control boxes of legacy robots to shrink down to a small NUC-sized device mounted directly on the robot. (00:18:05) The SO101 Robot Arm: Introducing the SO101 3D-printed robot arm from Hugging Face, which democratizes robotics by allowing anyone to build a leader-follower setup for around $200. (00:21:20) How “Backyard Engineers” Learn: Kevin’s advice for firmware and hardware engineers stepping into robotics: follow the documentation, get it running, and then ask questions about what you don’t know. (00:23:55) Calibration and the LeRobot Framework: A look at using the LeRobot open-source framework to calibrate hobby-grade motors and define their movement limits. (00:26:40) Recording “Episodes” via Imitation Learning: How users physically guide the leader arm to control the follower arm while a webcam records the visual and servo coordinate data. (00:28:50) Training the Model: Organizing data into short 5-episode chunks to make deletion easier, and training the model locally on NVIDIA GPUs or in the cloud. (00:31:30) From Training to Evaluation: Moving from training to evaluating the custom model, and asking questions about action chunking, model stutter, and operating frequency. (00:35:10) Sensor Fusion vs. Pure Vision: The current dominance of cameras in physical AI, and the potential to fuse accelerometers and time-of-flight sensors on mobile robots like Autonomous Mobile Robots (AMRs). (00:41:00) Real-World Calibration Challenges: Kevin shares a story of someone knocking over his camera boom at Hannover Messe and how he used April Tags to quickly recalibrate the camera’s physical coordinates. (00:43:15) The Reality of Humanoids: Debunking humanoid hype and explaining why full humanoids are further out than the public thinks, due to immense hardware cost, degrees of freedom, and safety. (00:45:50) The Puppeteer behind the Curtain: Kevin points out that many impressive humanoid demos, like the Unitree G1 at Hannover Messe, are actually being teleoperated by an engineer standing nearby. (00:48:40) The Complexity of Robot Subsystems: Using Steve Crunch and the book “Exploding the Phone” as an analogy for how modern robotic systems have become too complex for a single human to fully understand. (00:52:15) Understanding SmolVLA and SmolVLM: Diving under the hood of Vision Language Action models, which are fine-tuned transformer models that translate visual inputs into physical robotic coordinates. (00:55:10) Local AI and the NPU: How modern System-on-Chips (SOCs) let engineers run models locally on the CPU, GPU, or Neural Processing Unit (NPU) using optimization toolkits like Intel OpenVINO. (00:58:15) The Joy of the Physical World: Why working with physical hardware and robots is far more creative and rewarding than optimizing spreadsheets or SaaS applications. (01:04:30) “If the Robot Can’t Kill You, It’s Not Fun”: Kevin shares his colleague’s favorite metric for a truly exciting robotics project. (01:07:10) Advice on ROS and “Cobbling”: Starting with duct tape and bubble gum, and learning message-passing frameworks like ROS only when you need to coordinate multiple independent robots. (01:09:30) Physical AI Studio Demo: An invitation to see Kevin showcase Intel’s Physical AI Studio with multiple active robots at the upcoming AI Infra conference in Santa Clara. (01:14:45) Find Kevin online at his website cloutier.engineer or on LinkedIn. (01:20:10)