atomiq bg mobile
atomiq bg desktopatomiq kv desktop scaled

Atomiq

Ambitious Edge AI

atomiq kv mobile

The Atomiq™ SoC enables a new class of always-on edge AI through SPOT®, delivering high-performance NPU workloads with unmatched power efficiency. Targeted applications include advanced audio, natural language processing, computer vision, and on-device reasoning for intelligent, responsive, and energy-efficient edge applications.

High Performance Meets Power Efficiency

January 6, 2026

The World’s First NPU SoC Built on SPOT

01

AI in Overdrive

Atomiq leverages the Arm® Ethos™-U85 NPU to deliver over 200 GOPS of on-device AI performance, with support for sparsity and on-the-fly weight decompression to accelerate complex workloads.

02

Dynamic Power Scaling

SPOT-based ultra-wide range dynamic voltage and frequency scaling (DVFS) enables operation at lower voltages and power than previously possible, making advanced, always-on edge AI practical.

03

Robust Software

Ambiq’s helia™ AI platform, ADKs, and neuralSPOT® SDK deliver a tightly integrated hardware-software stack that enables high performance, lower memory footprint, ultra-low power, and significantly faster AI deployment.

04

Leading Energy-Efficiency

The SPOT platform brings breakthrough ultra-low-power efficiency, dramatically reducing energy consumption for next-generation edge workloads.

March 5, 2026

12nm SPOT Platform Redesigned

01

TSMC 12nm FinFET Node

Atomiq is built on TSMC’s 12nm FinFET node, which, when combined with the SPOT platform, enables high-speed sub- and near-threshold operation that was previously impractical in earlier planar nodes.

02

Power Levels as Low as 300mV

The new Ultra-Low Power (ULP) mode is engineered to operate close to 300mV while maintaining practical performance levels, expanding Atomiq’s dynamic range across ULP, Low-Power (LP), and High-Performance (HP) modes.

03

Multi-Channel SIMO Buck Converter

Featuring a multi-channel Single-Inductor Multiple-Output (SIMO) buck converter that adapts across processes, temperatures, and workloads, delivering 6 different voltage rails at loads ranging from nanoamps to milliamps.

04

Purpose-Built for Edge AI

By performing hundreds of multiply-accumulate (MAC) operations in parallel, Atomiq achieves tens of billions of operations per second at clock speeds as low as 100 MHz, enabling edge devices to process complex AI models

Coming Soon…

Memory Efficiency Like No Other

01

02

03

04

Frequently Asked Questions (FAQ)

  • Atomiq™ is Ambiq’s next-generation SoC architecture integrating an NPU, designed to enable scalable, ultra-low-power AI at the edge. It’s built on our patented SPOT® platform and represents a major milestone in how we architect future always-on AI systems.

  • Apollo is the foundation of Ambiq’s current edge AI success. Atomiq™ extends that foundation, introducing a new system architecture optimized for higher-performance, always-on AI workloads as models grow larger and more complex.
    A simple way to think about it:
    Apollo powers today’s edge AI deployments; Atomiq defines the next phase.

  • Absolutely. Apollo remains central to our portfolio and revenue today. Atomiq™ builds on Apollo’s success—it doesn’t replace it.

  • Atomiq™ introduces a system-level architecture designed from the ground up for AI workloads, including a dedicated NPU, while retaining Ambiq’s leadership in ultra-low-power operation through SPOT®. It’s about scaling intelligence without scaling power.

  • Arm® Ethos™-U85 provides meaningful gains in AI performance and efficiency, and it aligns well with our SPOT® platform. Together, they enable larger models and more complex workloads at power levels that weren’t previously practical at the edge.

  • Applications that demand always-on intelligence, such as:

    • Conversational AR glasses
    • Advanced wearables
    • Smart cameras

  • They represent two ends of the edge AI spectrum—ultra-compact consumer devices and mission-critical industrial systems. Together, they show how Ambiq’s platforms scale across vastly different constraints and use cases.

  • We haven’t announced availability timelines. We plan to share additional technical details and roadmap updates beginning in March 2026 at Embedded World.

Preparing to download