OmniVision has introduced the OAX7700, an ASIC the company identifies as its lowest-power AI companion processor built for always-on vehicle security systems. The part integrates a neural processing unit (NPU) alongside its image signal processing path, and OmniVision is positioning it explicitly as an upgrade path for existing exterior camera installations — with surround-view setups called out as the primary candidate for replacement.

What does the OAX7700 change for automotive camera design?

The commercial logic behind the device is power draw. Always-on vehicle security — the kind that monitors a parked car's perimeter and cabin through exterior cameras — only works if the silicon sipping from the vehicle's battery can survive long dwell times without draining the 12V system. By claiming the lowest power envelope in its AI companion processor line, OmniVision is arguing that automakers can now run continuous, NPU-accelerated vision workloads on camera nodes that today perform simpler, less capable processing.

The target installation base matters here. Surround-view systems — the multi-camera rigs that stitch a bird's-eye image for parking and low-speed maneuvering — are already standard fitment across large portions of the new-vehicle fleet. Those cameras exist; their image sensors, in many cases, are already OmniVision parts. The OAX7700 is designed to slot in as a companion processor alongside that sensor base, adding on-device AI inference for security functions such as intrusion or impact detection without forcing a redesign of the camera modules themselves.

That companion positioning distinguishes the OAX7700 from the main automotive vision SoCs that drive ADAS stacks. OmniVision frames it as an add-on companion device: the domain controller handles driving functions, while the OAX7700 handles the always-on monitoring workload that would otherwise keep a much larger, hungrier processor awake.

Where does the NPU fit?

The integrated NPU is the enabling element for the ultralow-power claim. Rather than streaming raw or lightly processed video to a central compute unit — an approach that keeps high-bandwidth links and power-hungry processors active around the clock — the OAX7700 runs inference locally at the camera node. Event detection, classification, and the decision about what is worth waking the rest of the vehicle's electronics for can all happen on the ASIC itself.

This is the same architectural pattern that has reshaped edge vision across consumer and industrial markets over the past several years: push inference to the sensor, cut the data that has to move, and cut the power budget with it. Automotive is a harder version of the problem because of temperature ranges, AEC qualification requirements, and the unforgiving arithmetic of battery drain over days or weeks of parking. OmniVision's bet is that its low-power claim clears that arithmetic where earlier companion processors did not.

Why target surround-view specifically?

Surround-view is the anchor use case because the cameras are already there. An upgrade that converts existing perimeter cameras into an always-on security mesh requires no new glass on the vehicle, no new mounting points, and minimal change to harnessing — the value-add lands almost entirely in the companion silicon and its software. For tier-one suppliers and OEM platform teams, that is the cheapest route from a passive parking aid to an active security feature that can be marketed at retail as a differentiator.

The security application space the OAX7700 addresses — monitoring for break-ins, towing, vandalism, and collision events while the vehicle is parked — is one where automakers have been searching for implementations that do not impose a meaningful range penalty on electric vehicles. A companion processor that draws milliwatts rather than watts changes that trade-off directly.

What comes next?

OmniVision has framed the OAX7700 around retrofit-ready exterior camera upgrades, starting with surround-view; if the power claim holds in production vehicles, expect the companion-processor pattern to spread across cabin-facing and rearview camera nodes as well, tightening competition in the automotive edge-AI silicon segment.