Vellex Logo Golden

OUR TECHNOLOGY

Computing with Physics

The Natural Path from Signal to Insight

Sidestepping the Digital Bottleneck

At Vellex, we don't just accelerate digital computation—we sidestep its most inefficient steps by harnessing the laws of physics

Conventional digital computing requires forcing the continuous, analog nature of the physical world into a discrete grid of ones and zeros. This translation creates inherent latency and energy waste. Our approach is fundamentally different: instead of translating the world for a computer, we build a computer that speaks the language of the physical world.

Brute Force vs. Physics

image

Standard computing forces you to discretize data into a grid and use iterative, step-by-step algorithms to search for an answer. This "brute-force" approach consumes significant time and energy.

Instead, the Vellex approach is to map the problem onto a physical circuit. Like a ball naturally rolling to the bottom of a curved surface, our system uses the laws of physics (like the principle of minimum energy) to settle on the optimal solution instantly.

image

From Physical Principle to Programmable IP

Our programmable analog circuits act as a dynamic canvas where raw sensor data continuously shapes the circuit’s 'energy landscape.' The solution — whether an optimal control parameter, a key AI feature, or a fault signature — emerges as a stable voltage or current. The only data that needs to be digitized is this final, high-value result.

Grounded in Research

Invented by our co-founders at Stanford University and exclusively licensed to Vellex, this hybrid analog-digital approach is proven, not theoretical. Built on published research and validated through real-world applications, our technology solves complex optimization and control problems with remarkable speed and efficiency.

The result is computation that occurs through the near-instantaneous relaxation of a physical system. This provides orders-of-magnitude improvements in speed and power efficiency — feeding AI models cleaner data faster, and enabling truly real-time control without the data overload.

Technical Advantages

image

Our physics-based approach drives a fundamental shift in intelligent system design. By moving computation to the analog front-end, we deliver system-wide gains in power efficiency, latency, and performance.

  • image

    Drastically Reduced Digitization Overhead

    Our analog-first approach performs heavy computations — such as filtering and feature extraction — before the signal is ever digitized. This significantly lowers the required sampling rate and bit depth for Analog-to-Digital Converters (ADCs), or allows for ADC-free architectures where only the final result is converted. This eliminates the 'digitization tax,' slashing power and complexity at the sensor edge.

  • image

    Intelligent Data Reduction at the Source

    Instead of logging raw noise, our IP computes on signals in real-time to extract only the salient information. In video quality control, for example, we can reduce 30MB of raw image data to just 3KB of defect coordinates. This >99% reduction means AI models are fed cleaner data, training datasets become leaner, and inference becomes faster.

  • image

    Real-Time Action Over Low Bandwidth

    By transmitting only compact, high-value insights instead of raw streams, we slash bandwidth requirements by up to 95%. This unlocks sophisticated real-time control for robotics and sensor networks operating on low-power connections (like LoRaWAN or NB-IoT), where transmitting raw data would be impossible.

Our Technology: Integration & Partnership

image

Flexible & Scalable: Our IP Integration Model

Our goal is to make physics-based computing accessible to system designers through a flexible semiconductor IP model that complements, rather than replaces, your existing digital architecture.

  • image

    Programmable Analog IP Blocks

    Our primary product is a licensable IP block designed to work in tandem with your existing digital processors (CPU or MCU). It acts as a specialized co-processor to handle the heaviest math at the sensor front-end.

    Seamless Integration: The digital processor remains in control, configuring the analog block and receiving final results.

    Offload the Heavy Lifting: Handles complex pre-processing and feature extraction in the analog domain.

    Empower the Edge: Frees up the digital core to run more complex AI models, like lightweight CNNs or decision algorithms .

  • image

    Fully Integrated Sensor SoCs

    For high-volume, dedicated applications, we can tightly integrate our IP with sensors and digital logic to create a complete System-on-Chip (SoC). This offers the ultimate in miniaturization and efficiency.

    Smart Sensors: Enables on-device AI for applications like continuous health monitoring or machine vision.

    Data Minimization: Processes raw data locally and sends only critical alerts or metadata to the cloud.

    Partnership Model: We co-develop these next-generation components with industry leaders in sensing and processing.

Learn More

image
San Francisco, CA
+1 (415)-212-8421contact@vellex.ai
Whether you're a potential partner, investor, vendor, or simply curious, we'd love to keep in touch.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
By submitting your email address, you agree to our Privacy Policy and provide consent to receiving updates from our company.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.

Quick Links

image
HomeTechnologyAbout usNewsroom
Join us

Follow us on

image
image
@vellex.ai
© 2025 Vellex Computing. All Rights Reserved.
Privacy PolicyTerms of ServiceCookies Settings