High-Precision In-Sensor Computing Reaching Up to 10 Bits

The growing demand for data-centric and intelligent applications calls for seamless integration of sensing, memory, and computation within a single device to improve efficiency, reduce energy consumption, and enable real-time processing. However, the adoption of such in-sensor computing is hindered by limited reconfigurability, which constrains its use in complex tasks beyond basic image processing. Here, we identify cascaded switching driven by interactions among differently polarized domains as a fundamental barrier to achieving high reconfiguration precision. To overcome this, we introduce a polarization energy focusing strategy, enabled by a tailored compositional distribution of ferroelectric and semiconductor materials. This approach yields the first in-sensor computing device with reconfigurable precision reaching 10 bits. Its capability in both conventional image processing and advanced reconstructed optics is demonstrated, highlighting its potential as a high-linearity, high-precision platform for next-generation intelligent systems.