Sony Patents a System That Turns Off AI Chip Memory the Moment It's Done With It
Most AI chips keep their memory fully powered even when the data inside has already been used. Sony's new patent describes a smarter bookkeeping system that shuts down those idle memory sections the instant they're no longer needed.
How Sony's chip watches memory and cuts its own power
Ever tried to keep every light in your house on just because you might walk into one of the rooms? That's roughly what AI chips do with memory right now: they keep every section powered up, even after the data inside has already been read and processed.
Sony's patent describes a way to fix that. As the chip processes data for an AI task (like recognizing faces in a camera feed), a small tracking layer watches which sections of memory have had all their contents read. The moment a section is fully used, the power management system either turns that section off completely or drops it into a low-power standby mode.
Sony says this approach targets sensors, including image sensors, which run AI tasks constantly. Shaving power from the memory chip doesn't sound exciting, but for a camera that runs 24 hours a day, it adds up fast.
… a power management unit that brings power to the memory bank where all the pieces of data stored in each memory word have been used from an on state into an off state or a low-power state where the memory banks operate with reduced power compared to the on state on a basis of the usage status information that has been retained.
Translation: A power controller shuts down memory sections the instant their data is no longer needed.
How the holding unit tracks every word and signals shutdown
The patent centers on memory that is split into separate blocks called memory banks. Each bank holds a portion of the data that a DNN (deep neural network, the type of AI model used for image recognition, object detection, and similar tasks) needs during processing.
A small dedicated circuit called a holding unit sits alongside each bank. It keeps track of usage status information, a running record of whether each individual memory word inside that bank has been read by the processing unit. Think of it as a checklist: every time the processor reads a piece of data, the holding unit ticks that item off.
Once every item on the checklist for a given bank is ticked, a power management unit takes over. It transitions that bank from its normal on state to either a full power-off or a reduced-power standby mode. The bank stays dark until it is needed again.
- Memory divided into independently switchable banks
- Per-bank holding unit tracks read status of every data word
- Power manager cuts or reduces power as soon as a bank is fully consumed
- Designed to work inside sensor systems, including image sensors running continuous AI inference
… retains usage status information indicating a usage status of the data stored in each memory word …
Translation: Small tracking units monitor whether individual pieces of data have been read yet.
What this means for AI cameras and always-on sensors
Battery life and heat output are the two biggest constraints on AI chips built into cameras and sensors. Memory is a significant chunk of a chip's power draw, and it's usually on constantly. A system that actively retires memory sections during processing could reduce that draw without slowing the AI down or changing what it outputs.
Sony's pattern of sensor-chip power-efficiency filings points toward image sensors that run AI at the edge, meaning on the device itself rather than in the cloud. Keeping those sensors cool and power-efficient is important for everything from security cameras to autonomous vehicles, where the chip runs nonstop and heat or battery drain can become a real design ceiling.
Sony's seventh filing we've tracked since July in the AI chip wars follows daisy-chained display chips and parallel image compression.
Always-on sensors and AI chips burn power continuously, and a surprising share of that waste comes from memory staying fully powered even after it has finished its job. In a camera, a wearable, or any small battery-powered device, that idle draw shortens lifespan, raises heat, and forces tradeoffs in how compact the hardware can be.
Sony's approach tracks, at a fine-grained level, when each section of memory has delivered its data to the AI process and then cuts power to that section immediately. Millions of always-on devices face exactly this drain every day, which makes the problem large enough that even modest improvements per device accumulate into something real at scale.
How much difference this makes in practice will depend on how deeply it gets built into actual products, but the cost of idle memory is measurable and persistent, and an approach that addresses it directly is aimed at the right target.
There are more where this came from
We read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.
The drawings
14 drawing sheets from US 2026/0259594 A1 · click any drawing to enlarge
Want this weekly breakdown for a company we don't cover? Patentlyze Pro →