🧠 What if your car's AI chip worked like a human brain? Honda is teaming up with U.S. startup Mythic to co-develop a "neuromorphic SoC", a chip inspired by how our brains think. Said to rival NVIDIA's GPUs in power efficiency, this technology could redefine autonomous driving.

Why Honda Is Betting on Brain-Inspired Chips

On February 4, 2026, Honda Motor Co. announced an investment in Mythic, a Texas-based AI semiconductor startup, along with plans for Honda R&D Co., Ltd. to co-develop next-generation automotive system-on-chips (SoCs) with the company.

The timing reflects an urgency within Honda's strategy. The automotive world is rapidly shifting toward SDVs, software-defined vehicles, cars whose capabilities are defined by software and can continuously evolve through updates even after purchase. To power this vision, Honda needs SoCs with an order-of-magnitude leap in computing performance. Everything from autonomous driving and advanced driver-assistance systems (ADAS) to powertrain control and comfort features will be managed by a single, centralized electronic control unit (ECU).

But there's a fundamental problem: boosting AI computing power nearly always means consuming proportionally more energy. For a vehicle, especially an EV with limited battery capacity, that's a critical tradeoff. Honda recognized that solving this "performance vs. power consumption" dilemma requires looking beyond conventional digital computing. Their answer: neuromorphic technology.

What Exactly Is a Neuromorphic SoC?

"Neuromorphic" literally means "shaped like a nerve." It refers to semiconductor technology inspired by how the human brain processes information.

Standard computers operate on what's called von Neumann architecture, where the processor (CPU) and memory are physically separate. Every computation requires shuttling data back and forth between them, a process that creates delays and guzzles power. This well-known limitation is called the "von Neumann bottleneck."

The human brain works entirely differently. Neurons and synapses store and process information simultaneously, all in one integrated system. Our brains run on roughly 20 watts, less than a household light bulb, yet they handle pattern recognition, prediction, and decision-making far more elegantly than any supercomputer.

A neuromorphic SoC attempts to replicate this brain-like integration in silicon. By combining computation and memory in the same place, it drastically reduces the data transfer overhead that plagues conventional chips, and the enormous power consumption that comes with it.

Mythic's Secret Weapon: Analog Compute-in-Memory

Mythic was founded in 2012 as a spinoff from the University of Michigan and is headquartered in Austin, Texas and Redwood City, California. The company is backed by venture capital firms including Lux Capital and DCVC, and in December 2025 it closed a roughly $125 million funding round whose investors included Honda and Lockheed Martin.

At the heart of Mythic's technology is "analog CiM" (Compute-in-Memory). While conventional AI chips process information as digital signals (ones and zeros), Mythic performs calculations directly inside flash memory cells using analog signals. Memory elements act as tunable resistors: inputs arrive as voltages, and outputs emerge as electrical currents. This means multiply-and-accumulate operations, the mathematical backbone of AI, happen right where the data is stored.

The advantage is straightforward: since data doesn't need to travel between memory and processor, the power consumed by data movement drops dramatically. Mythic's M1076 Analog Matrix Processor (AMP) reportedly delivers up to 25 TOPS (25 trillion operations per second) at just 3–4 watts of power.

For perspective, high-performance GPUs typically consume tens to hundreds of watts. Some observers have noted that Mythic's chips offer a compelling advantage over NVIDIA GPUs in terms of performance per watt, a metric that matters enormously in vehicles.

Honda's Two-Pronged Semiconductor Strategy

To fully appreciate the Mythic partnership, you need to see it alongside Honda's other major chip collaboration. In January 2025 at CES, Honda announced a partnership with Renesas Electronics to develop an SoC targeting 2,000 TOPS of AI performance at 20 TOPS/W power efficiency, built on TSMC's cutting-edge 3nm process. That chip, combining Renesas's 5th-generation R-Car X5 SoC with Honda's custom AI accelerator via chiplet technology, was meant to power the "Honda 0 (Zero) Series" EVs launching in the late 2020s, representing the leading edge of digital computing.

That plan, however, has been thrown into doubt. On March 12, 2026, citing sharply deteriorating finances, Honda announced a sweeping reversal of its EV strategy, cancelling three North American EVs, the Honda 0 SUV, Honda 0 Saloon and Acura RSX, and flagging losses of up to ¥2.5 trillion (about $15.7 billion) (details). With the 0 Series shelved, the future of the Renesas SoC that was meant to power it is now uncertain.

The Mythic partnership, by contrast, looks further ahead. It's Honda's hedge against the eventual limits of digital scaling, a bet on analog-based neuromorphic computing as a fundamentally different approach that could deliver the next breakthrough. Even with the Renesas plan in flux after the EV retreat, Honda's decision to keep working with Mythic shows it is still placing long-term bets on making in-car AI far more power-efficient.

Why Power Efficiency Is Life or Death for Autonomous Driving

Consider what happens inside a Level 3+ autonomous vehicle. Multiple cameras, LiDAR sensors, and millimeter-wave radar generate enormous volumes of data every second. The AI must perform object detection, path prediction, and driving decisions in real time. Current ADAS chips deliver hundreds of TOPS but frequently consume over 100 watts.

For an EV, that's power directly subtracted from driving range. If the AI system continuously draws 100 watts, it meaningfully reduces how far the car can travel on a single charge. Heat is another problem, high temperatures degrade semiconductor reliability and may require additional cooling systems, adding cost and weight.

A neuromorphic SoC achieving equivalent performance at a fraction of the power could fundamentally change this equation. It would allow sophisticated autonomous driving features without significantly compromising range, a genuine game-changer for the EV era.

The Global Neuromorphic Race

Honda and Mythic aren't operating in a vacuum. A global competition in neuromorphic technology is intensifying. Intel's "Loihi" series pursues real-time processing through spiking neural networks. IBM's "TrueNorth" famously simulated one million neurons on a single chip. BrainChip's "Akida" targets ultra-low-power edge AI. In Europe, the Fraunhofer Institute is advancing next-generation neuromorphic chips using ferroelectric transistors.

What makes Honda's involvement distinctive is that it's an automaker directly participating in chip development, not a semiconductor company or tech giant. By embedding automotive requirements (vibration resistance, temperature extremes, safety certifications) into chip design from the very beginning, Honda can ensure that the resulting technology is purpose-built for vehicles rather than adapted from other applications.

The Road Ahead: Cars as Rolling AI Platforms

The Honda-Mythic partnership is more than a procurement story. It's an attempt to replicate the human brain's architecture in silicon and deploy it in one of the most demanding environments imaginable, a moving vehicle.

Pressing ahead with Mythic even as it overhauls its EV strategy, Honda is signaling how traditional automakers might try to stay competitive in the AI era. As cars transform from simple transportation into rolling AI platforms, brain-inspired chips may fundamentally reinvent what a vehicle's "brain" can do.

In Japan, Honda's bold move into neuromorphic semiconductor development is generating significant attention. How is your country approaching the challenge of energy-efficient AI for autonomous driving? Are your domestic automakers getting involved in chip development? We'd love to hear your perspective.

References