Silicon To Software | Autonomous Systems | Cybersecurity

Silicon To Software | Autonomous Systems | Cybersecurity

By Imran ValianiTechnology
Download on the App Store

All Silicon To Software | Autonomous Systems | Cybersecurity episodes

  • Toyota Solid-State Battery Timeline: Why the 2027–28 Launch Depends on Plants Still Under Construction

    Toyota's solid-state battery has a launch target of 2027. One of its critical materials will come from a plant that's still under construction — and that construction is scheduled to finish in 2027 too. That's the part of the story most headlines skip.

    Headlines say 2027. Toyota's own releases say 2027–28. And in November 2026, Toyota described the goal as a future model by the end of 2028. But a launch isn't volume. When Toyota and Idemitsu announced their partnership in 2023, they described two stages: first commercialization, then full-scale mass production. Most headlines kept stage one and dropped stage two.

    This episode walks through the four supply layers under Toyota's battery: lithium sulfide from Idemitsu's new Chiba plant (construction due to finish June 2027, sized at the equivalent of 3 GWh of batteries a year); the solid electrolyte itself, where a verification plant making more than a dozen tons of samples a year steps up to a several-hundred-ton pilot in 2027, with full mass production — several thousand tonnes, several tens of thousands of vehicles — put at 2030 and beyond; the cathode, where Sumitomo Metal Mining's durable material enters mass production as early as the financial year beginning April 2028, the latest date in the chain; and cell production, where the widely quoted "2026, 9 GWh" belongs to a liquid lithium-ion battery — on the solid-state row, the start date and capacity are blank.

    Toyota isn't late — its own leadership said in 2023 that solid-state volumes would stay small in 2027. The point is narrower: when the headline says solid-state is here, it'll mean something smaller than most people picture. Skip the launch event; watch the plants.

    For the full written analysis, sources and diagrams, head to silicontosoftware.com.

    Subscribe and follow Silicon to Software for more breakdowns of the hardware behind AI, EVs and batteries — from power shelves to PCB laminates.

    Chapters:

    00:00 The 2027 launch — and the plant still being built

    00:27 Headlines say 2027 — Toyota says 2027–28

    00:42 A launch isn't volume

    00:51 Two stages: commercialization, then mass production

    01:03 The laminate analogy: volume belongs to the supplier

    01:27 Layer one: lithium sulfide, from an oil refiner

    02:07 Layer two: the solid electrolyte ladder

    02:48 Layer three: the cathode — April 2028 at the earliest

    03:17 Layer four: the cells and the METI mix-up

    03:40 The fair pushback: Toyota isn't late

    04:09 Skip the launch event, watch the plants

    04:39 Subscribe and follow

    5 min
  • Battery Storage for AI Data Centers: 3 Critical Risks in the Power Path

    On the evening of July 10, 2024, a transmission line in Northern Virginia faulted, and about 1,500 MW of load dropped off the grid. None of it was homes. None of it was factories. All of it was data center load. It has happened twice more in the same zone — most recently on July 22, 2026, when PJM put the load reduction at 3,800–3,900 MW, the largest of its kind in PJM's history.

    AI training makes the problem worse. In a normal data center, thousands of unrelated jobs average each other out. AI training doesn't work that way — the GPUs compute together and pause together, so their power curves stack instead of canceling. NVIDIA says a single rack can swing from about 30% of its power draw to full power and back in milliseconds.

    So where does storage actually sit? In three layers: capacitors in the GB300 power shelf (about half a power supply by volume, 65 joules per GPU — a shock absorber measured in hundredths of a second, not backup power), lithium-ion battery backup units in the rack, inches from the IT load, and big battery systems at the facility level, where the site connects to the utility.

    Then the three risks most people pricing this storage aren't seeing: cycle life (smoothing storage cycles all day, and electrolytic capacitors hate heat — every 10°C cooler roughly doubles their life), fire (the September 2025 fire in the lithium-ion UPS batteries at South Korea's National Government Data Center damaged 96 of 647 government systems and destroyed all 384 UPS battery units), and the board itself (NVIDIA is targeting 2027 to move to 800 VDC — 1 MW is about 1,250 A at 800 V versus 18,500 A at 54 V — but spacing gets harder, laminate choice matters more, and conductive anodic filament becomes a real reliability risk under years of DC bias).

    That leaves one question for anyone specifying this storage: if it smooths the workload all day, does it still have enough charge left to ride through a grid fault?

    For the full written analysis, the sources and the diagrams, head over to silicontosoftware.com.

    Subscribe and follow Silicon to Software for more breakdowns of the hardware behind AI — from power shelves to PCB laminates.

    Chapters:

    00:00 The Virginia event: 1,500 MW drops off

    00:26 It's happened twice more

    00:47 Where the storage actually sits

    01:08 The workload: GPUs pause together

    01:41 Layer one: capacitors in the rack

    02:24 Layers two and three: BBUs and facility BESS

    02:40 Risk one: cycle life

    03:16 Risk two: fire

    03:52 Risk three: the board

    04:32 The question to ask

    5 min
  • Direct-to-Chip Liquid Cooling in AI Racks: Where the Heat Problem Really Went

    Liquid cooling didn't remove the heat problem from AI racks. It moved it into the board.

    In late 2024, reports said Blackwell GPUs were overheating in 72-processor servers, and suppliers spent months working through overheating and coolant leaks in GB200 liquid-cooling systems. The root causes were never disclosed — but the reports tell you something: a liquid-cooled rack needs a lot more than raw heat transfer to work.

    In this episode: how the loop actually works — a cold plate on each GPU package, coolant through quick disconnects into a rack manifold, then a CDU handing heat to the building's loop. Why today's GB300 racks are still hybrid, about 85% liquid, with optics, storage and some power parts still on air. And why HPE's three rack-power numbers — 132 kW nominal, about 155 kW peak, 192 kW provisioning — answer three different questions.

    Then the part most people miss: the cold plate has to press on the package, hard and evenly, so the TIM makes good contact. A 2006 study of one thermal interface material found thermal resistance could differ by up to 45% between high- and low-pressure application. That clamp force is carried through the package into the solder joints, into the board and the backplate. Too much force stresses the BGA and bends the board; too little and you get hot spots. There's a window, and it has to be qualified for each assembly.

    Plus: Rubin goes to 100% liquid with 45°C coolant, PG25 chemistry and quick-disconnect leak risk, liquid-cooled busbars up to 5,000 amps, and Microsoft's lab microfluidics cooling up to three times better than cold plates.

    Liquid can carry the heat. Whether the rack stays reliable gets decided at the boundaries between suppliers — and the board sits right in the middle.

    Full written analysis with figures and every source: https://www.silicontosoftware.com/direct-to-chip-liquid-cooling-ai-racks/

    Subscribe and follow Silicon to Software for more breakdowns of where AI hardware actually breaks.

    Chapters:

    00:00 Welcome to Silicon to Software

    00:18 The Blackwell overheating reports

    00:55 How the loop works

    01:10 Hybrid racks and three power numbers

    01:42 Clamp force and the TIM

    02:59 Rubin goes 100% liquid

    03:29 The 45°C coolant trade-off

    03:50 PG25, leaks and liquid-cooled busbars

    04:14 Microfluidics: cooling the silicon itself

    04:26 What it means for board builders

    6 min
  • Tesla 4680 Dry Electrode Process: Why the Cathode Took Years, and What's Still Unproven

    Back in 2021, Tesla was trying to ramp its 4680 cell — and one of the first signs of trouble didn't come from a chemistry lab. It came from the steel rollers. The cathode material was so hard, it was damaging the calender rolls — and only when the rolls got bigger.

    In this episode:

    - What a dry electrode actually is — and why skipping the solvent slurry moves the control problem from the drying ovens to the mixer and the calender

    - Why the cathode, not the anode, became the bottleneck — and the reports of Tesla losing most of its cathodes in test production

    - The PTFE binder trade-off: it holds the film together, but blocks electrons and ions and doesn't always stick to the foil

    - The defect that worries a manufacturing guy most: failures that don't show up until months later

    - The January 2026 update: Tesla says it's making 4680 cells with both electrodes dry-processed in Austin — and what still isn't published (yield, scrap rates, cost per kWh)

    - The real test: copy exactly — same yield on every line, every site, every shift. A process isn't solved when it works. It's solved when it's boring.

    Full analysis: https://www.silicontosoftware.com/tesla-4680-dry-electrode/

    5 min
  • Glass Substrates for AI Chips: Why Mass Production Keeps Slipping

    Every high-end AI processor sits on a package substrate — a tiny, dense wiring board connecting the chip to the motherboard. For decades its core has been organic, a woven glass-fiber laminate. AI packages are getting huge, and at that size a familiar problem gets loud: warpage.

    In this episode:

    - Why warpage threatens assembly yield: silicon, the organic core, and copper expand at different rates when heated, so the stack bends — and Intel has shown an engineering sample carrying about twice the silicon a lithography tool can print in one shot

    - What glass is supposed to fix: it is stiff, extremely flat, and its expansion can be tuned closer to silicon

    - Measured results, with their limits: test vehicles from Amkor and STATS ChipPAC showed roughly a third less warpage with a glass core — specific test designs, not a general rule; in one study, packages with no protective edge coating failed and coated ones passed

    - The simulation that complicates the story: a Unimicron model, reported secondhand and not a physical test, found glass roughly halved strain where the chip meets the substrate but more than doubled strain on the larger solder joints where the substrate meets the board

    - Why glass cracks: copper vias run through the glass and copper expands far more than glass, with cracks pointed to at via walls, via caps, and panel cut edges

    - Where the big players stand: Absolics (SKC's subsidiary in Georgia) in an initial reliability evaluation with samples in Taiwan, with a delay acknowledged in SKC's own financing materials; Samsung Electro-Mechanics with two separate targets — the glass core in the second half of 2027 and finished substrates from 2028, with a disputed claim about a failed prototype; and Intel, which put glass on its roadmap in 2023 and has shown engineering work but has named no customer or product, with no revised timeline found

    - The five gates: customer reliability testing is only the first — then process-window mapping, holding it lot after lot, sustained yield, and costs that compete with mature organic substrates; no glass-core supplier has published production yield data

    The verdict: a flat substrate is a materials result. A qualified substrate is a manufacturing result. Next time you see a glass announcement, ask which of the five gates it actually clears.

    Full analysis: https://www.silicontosoftware.com/glass-substrates-for-ai-chips/

    5 min
  • NVIDIA 800V DC: The Power Problem It Doesn't Solve

    A one-megawatt AI rack fed at 54 volts has to pull about 18,500 amps. At 800 volts, the same rack draws about 1,250. NVIDIA estimates a 54-volt megawatt rack would need up to 200 kilograms of copper busbar, and its power shelves alone would eat up to 64 rack units.

    In this episode:

    - Why rack power is climbing toward a megawatt: more GPUs packed into one rack so they can talk over short copper NVLink links

    - The 54-volt comparison: roughly 15 times less current at 800 volts, far less heat in the copper

    - The comparison most coverage misses: against 480-volt AC already in many data halls, each wire carries almost the same current — the gain is two conductors instead of three

    - The transitional rack: the first 800-volt hardware converts back down to 54 volts inside the compute rack, adding a conversion step rather than removing one; NVIDIA projects up to 5% better efficiency, and as of late October no independent measurements from an operating 800-volt AI data hall were found

    - Where the hard engineering moves: converters close to the GPU (silicon carbide, gallium nitride, stacked topologies — vendor numbers, not field data), and the circuit board, where 800-volt spacing, laminate choice, manufacturing tolerances, and cleanliness become safety issues

    - Protection and certification: DC arcs are harder to extinguish than AC arcs, grounding is still undecided, solid-state breakers are only mature at lower currents, and key standards were written around lower voltages or AC

    The verdict: against a 54-volt bus the physics is decisive, but 800-volt DC does not make the power problem disappear — it moves it into converters, boards, connectors, breakers, and certification.

    Full analysis: https://www.silicontosoftware.com/nvidia-800v-dc-architecture/

    5 min
  • Neuralink Mass Production: Can Its Brain Implants Really Scale?

    Can Neuralink actually mass-produce brain implants? On Dec 31, 2025, Elon Musk posted that Neuralink would start high-volume production in 2026, with surgery becoming almost entirely automated. That was a plan, not a production report. Texas filings show investment in office and manufacturing space near Austin, and job listings mention improving the implant manufacturing line — but no public production volumes or capacity numbers exist, which means nobody outside can measure scale yet. This episode walks through the engineering: the N1 implant's ~1,000 electrodes across 64 threads thinner than 5 microns, thin-film failure modes familiar to anyone who has built flex circuits, the moisture and hermetic-sealing challenge (no participant has carried the device for three years yet), the illustrative yield math behind 1,000 vs 3,000 channels, and the surgical throughput problem — a thousand implants on a shelf are not a thousand completed procedures. Full analysis: https://www.silicontosoftware.com/neuralink-mass-production/

    5 min
  • Sodium-Ion Batteries: Why CATL Bet $5 Billion on the Chemistry That Killed a Startup

    A California battery startup had working cells, customers, and a billion-dollar gigafactory announcement — then shut down in September 2025. Months later, CATL, the biggest battery maker on the planet, bet $5 billion on the exact same chemistry. Same chemistry, opposite bets. So who's right?

    In this episode:

    • Natron Energy's collapse — funding, certification delays, and a lithium price crash that wiped out sodium's cost story

    • What's actually different inside a sodium cell: hard carbon instead of graphite, aluminum foil on both sides

    • Energy density, honestly: CATL's 175 Wh/kg claim vs LFP and nickel-rich chemistries

    • Why sodium's raw-material cost isn't the same as finished-cell cost — and why nobody has shown cost parity yet

    • The cold-weather claim: ~90% retention at -40°C, and the unpublished test conditions behind it

    • Manufacturing reality: adapted lithium lines, hard carbon supply, and the scale problem

    The verdict: sodium-ion has made it from the lab to the factory — but it hasn't yet proven, with independent data, that it beats LFP on cost in a real car.

    Full analysis: https://www.silicontosoftware.com/sodium-ion-vs-lithium-ion-batteries-ev/

    6 min
  • Tesla AI5 Chip: Why Samsung and TSMC Make It So Complicated

    Tesla AI5 could be manufactured by Samsung and TSMC. But how can two different chip implementations run the same AI software?

    We break down Tesla AI5 chip manufacturing, semiconductor design, PCB power delivery, thermal management, and the hidden engineering costs of dual-foundry production.

    Tesla's AI5 strategy raises a difficult semiconductor engineering question: How do you maintain consistent software-visible behavior when two different foundries manufacture separate physical implementations of a processor?

    In this episode of Silicon to Software, host Imran Valiani examines the engineering challenges behind Tesla's proposed Samsung and TSMC manufacturing approach.

    With more than 20 years of experience in PCB manufacturing and technical sales, Imran explains why the problems don't stop at the silicon die.

    WHAT YOU'LL LEARN

    - Why Tesla is pursuing Samsung and TSMC for AI5 manufacturing

    - What changes when a semiconductor design moves between foundries

    - Why RTL and architecture can transfer while standard cells and physical layouts cannot simply be reused

    - How chip performance, power consumption, and operating limits may differ

    - Why BGA pin assignments, package compatibility, power sequencing, and PCB power delivery require validation

    - How two chips with similar total power can produce different thermal hotspots

    - Why formal equivalence, numerical equivalence, and real-time system correctness are separate verification objectives

    - What Tesla has publicly confirmed versus what remains unverified

    THE ENGINEERING PROBLEM

    A common logical architecture does not guarantee identical physical implementations.

    Samsung and TSMC use different manufacturing technologies, design rules, standard-cell libraries, and process-specific IP.

    Those differences can affect timing closure, power distribution, semiconductor packaging, thermal behavior, and system qualification.

    And even if both processors execute the same AI software, the complete system still has to meet its performance and timing requirements.

    For robotics and autonomous systems, producing the correct result is not enough. The result must also arrive when the system needs it.

    TESLA AI5 PRODUCTION OUTLOOK

    Tesla has discussed AI5 production plans for 2027, while Elon Musk has expressed hopes for volume production around mid-2027.

    Specific foundry process nodes, memory configurations, manufacturing allocations, and package-level compatibility details remain incompletely confirmed.

    This episode distinguishes reported manufacturing information from Tesla's public statements.

    READ THE FULL ENGINEERING ANALYSIS

    Read the complete article, including technical diagrams and supporting sources:

    https://www.silicontosoftware.com/

    ABOUT THE HOST

    Imran Valiani is the founder of Silicon to Software and has more than 20 years of experience in PCB electronics manufacturing and technical sales.

    Silicon to Software explores the physical engineering behind semiconductors, AI infrastructure, PCB manufacturing, embedded systems, and modern electronics.

    SUBSCRIBE

    Subscribe to Silicon to Software for technical discussions about the manufacturing constraints, hardware trade-offs, and reliability challenges behind today's technology.

    #TeslaAI5 #SemiconductorManufacturing #AIHardware

    5 min
  • Face ID vs Fingerprint Sensors: How They Really Work

    How does Face ID recognize your face, and how do fingerprint sensors actually work? Discover the infrared, ultrasonic, optical, and capacitive technology behind smartphone biometric authentication.

    From Apple's Face ID and Touch ID to Google's Pixel 8, we examine the engineering behind biometric security—and why these systems sometimes fail.

    In this episode of Silicon to Software, host Imran Valiani, with more than 20 years of experience in PCB electronics manufacturing, breaks down the hardware and physics behind the smartphone technologies we use every day.

    WHAT YOU'LL LEARN

    1. How Fingerprint Sensors Work

    Discover the differences between capacitive, optical, and ultrasonic fingerprint sensors.

    Learn how capacitive sensors detect changes caused by fingerprint ridges and valleys, how optical sensors use display illumination and reflected light, and how ultrasonic sensors measure acoustic echoes through the display stack.

    2. Why Samsung's Fingerprint Sensor Failed

    In 2019, Samsung acknowledged an unlocking issue affecting certain Galaxy smartphones using ultrasonic fingerprint recognition.

    We examine how textured silicone screen protectors created an unexpected security problem—and what the incident teaches engineers about sensor integration and accessory compatibility.

    3. How Apple's Face ID Works

    Apple's TrueDepth camera system projects invisible infrared dots to capture facial depth information.

    Learn how facial data is converted into mathematical representations and protected through the iPhone's Secure Enclave.

    We also explain why Face ID can operate in complete darkness.

    4. Apple Face ID vs. Google Pixel 8

    Google's Pixel 4 used dedicated infrared facial recognition hardware.

    The Pixel 8 took a different approach, using its conventional front camera and machine learning to achieve Android Class 3 biometric security.

    What does that mean for security, low-light performance, and hardware design?

    5. Can Biometric Authentication Be Fooled?

    Apple reports a random-person Face ID false-match probability of less than one in a million under its stated conditions.

    But random false matches and deliberate spoofing attacks are different security measurements.

    We explain why that distinction matters, how physical presentation attacks are evaluated, and why your passcode remains the ultimate fallback.

    VIDEO CHAPTERS

    [Insert Timestamps here]

    READ THE FULL ENGINEERING ANALYSIS

    For detailed technical explanations, engineering diagrams, and supporting research, visit:

    https://www.silicontosoftware.com

    ABOUT SILICON TO SOFTWARE

    Silicon to Software explores the engineering behind modern technology—from semiconductor devices and PCB manufacturing to embedded systems, automotive electronics, and AI hardware.

    Hosted by Imran Valiani, a technology professional with more than 20 years of experience in PCB electronics manufacturing and technical sales.

    CONNECT WITH SILICON TO SOFTWARE

    Website: https://www.silicontosoftware.com

    X: https://x.com/SiToSoftware

    Subscribe to the channel for engineering-focused technology explanations, hardware analysis, and semiconductor insights.

    #FaceID #FingerprintSensor #BiometricSecurity

    5 min