Amlogic launches two 6nm edge AI chips for cameras and industrial devices

Amlogic on Sept. 9 unveiled the C305X2 and A123X, two 6nm SoCs built around Arm’s Cortex-A320 CPU for battery-powered cameras, industrial vision and IoT devices. The chips are aimed at pushing more AI processing on-device, with samples available now and mass production planned for Q4 2026.

Why it matters: - Amlogic is targeting the fastest-growing part of edge AI: low-power devices that need local AI, video and security without constant cloud access. - The C305X2 and A123X are designed for battery-powered IP cameras, industrial machine vision, smart home cameras and other IoT devices that need more compute in less power. - The platforms combine on-device AI processing with hardware-grade Armv9 security, which matters for devices handling sensitive video and audio data. - The company says the chips can help OEMs and ODMs shorten product cycles and move legacy Armv8 designs to Armv9 with less code rework.

What happened: - Amlogic announced two new 6nm system-on-chips, the C305X2 and A123X, on Sept. 9 in Mountain View, California. - The chips use Arm Cortex-A320 CPUs and are among the first edge AI SoCs built on that ultra-efficient Armv9 core. - Samples are available now. - Mass production is scheduled for Q4 2026. - Amlogic said the unified platform is built for battery-powered IPCs, industrial machine vision, smart home cameras and multi-scenario IoT devices.

The details: - The Arm Cortex-A320 is described as the smallest and most power-efficient Armv9 CPU for IoT workloads. - Amlogic said the Cortex-A320 delivers a 34% SPECint2006 performance increase over the Armv8 Cortex-A35. - Amlogic said Cortex-A320 with SVE2 vector extensions provides about 10x ML processing uplift versus Cortex-A35. - Both chips pair the CPU with Amlogic’s in-house ADLA3 NPU rated at 8 TOPS for Transformer-native AI workloads. - The platform is designed to run lightweight computer vision models and compact LLMs locally. - Amlogic said local processing reduces cloud latency and keeps data on device. - The A123X uses TSMC 6nm process technology and a quad-core Cortex-A320 CPU. - The A123X adds a 4 TOPS ADLA2 NPU for object detection and tracking CNN models and an 8 TOPS ADLA3 NPU for Transformer workloads. - The A123X includes a neural-network sound event detection engine for low-power audio identification. - The A123X supports dual 4-lane MIPI image sensor inputs and a low-light HDR ISP. - The A123X supports up to 16M at 30fps in SDR and 13M at 30fps in HDR. - The A123X includes AI-ISP technology with support up to 4K at 30fps. - The A123X supports H.264 and H.265 encoding at 4K at 60fps and decoding at 4K at 30fps. - The A123X supports 32-bit LPDDR4, LPDDR4X and LPDDR5 memory up to 6400Mbps. - The A123X includes 2x4L MIPI CSI, BT.656/BT.1120, sub-LVDS 8-lane, MIPI DSI up to 1080P60, USB 3.0 OTG, USB 2.0 Host, Gigabit Ethernet and 2x CAN FD. - Amlogic positioned the A123X for machine vision, dashcams, sweeping robots, access control and video conferencing hardware. - The C305X2 also uses TSMC 6nm process technology and a dual-core Cortex-A320 CPU. - The C305X2 pairs an ADLA2 NPU, supporting AI-ISP algorithms up to 4K at 30fps, with an 8 TOPS ADLA3 NPU for Transformer workloads. - The C305X2 supports Always-on Video and Pre-roll low-power modes to extend battery life. - The C305X2 includes a neural-network sound event detection engine for low-power audio detection. - The C305X2 supports dual 4-lane MIPI image sensor inputs and a low-light HDR ISP. - The C305X2 supports up to 4K at 30fps with AI-ISP enhancement. - The C305X2 supports H.264 and H.265 encoding at 4K at 60fps and decoding at 4K at 30fps. - The C305X2 supports 32-bit LPDDR4, LPDDR4X and LPDDR5 memory up to 4266Mbps. - The C305X2 includes 2x4L MIPI CSI, BT.656/BT.1120, MIPI DSI up to 1080P60, USB 3.0 OTG, USB 2.0 Host and Gigabit Ethernet. - Amlogic said the C305X2 is optimized for solar doorbells, portable hunting cameras and compact 2K or 4K battery-powered IPCs. - Amlogic said the platform includes turnkey reference designs, pre-validated embedded OS stacks and unified NN SDKs compatible with mainstream ML frameworks. - The company said its Arm-based platform also includes production-ready firmware and global certification support. - The Amlogic platform is designed to support migration from older Armv8-based Amlogic chips. - Amlogic said the new SoCs are meant to help mainstream IoT devices balance AI performance, low power use and hardware-level data security.

Between the lines: - Amlogic is trying to position itself as a one-stop edge AI platform provider, not just a chip vendor. - The strategy centers on pairing a shared CPU foundation with different NPU and power profiles for industrial and battery-driven products. - The company is also leaning on Armv9 compatibility to make adoption easier for existing customers. - The focus on local inference, video codecs and low-light imaging shows Amlogic is chasing real device requirements, not just benchmark wins.

What's next: - Amlogic will move the C305X2 and A123X from samples into mass production in Q4 2026. - OEMs and ODMs are expected to use the reference designs and software stack to build next-generation edge AI devices. - The company and Arm are signaling broader rollout of Arm-based edge AI silicon across consumer and industrial markets.

The bottom line: - Amlogic is betting that edge AI hardware wins on the combination of local intelligence, low power and security, not raw compute alone. - The C305X2 and A123X are built to bring more AI work onto the device itself, where latency, privacy and battery life matter most.