PUBLISHER: ResearchInChina | PRODUCT CODE: 2109334
PUBLISHER: ResearchInChina | PRODUCT CODE: 2109334
Research on Autonomous Driving Sensor Chips: Deeply Perceiving the Physical World, Sensor Chips Are Playing A "Leading Role" in Intelligence
In 2026, the autonomous driving sensor chip industry hits a critical turning point. The access policy for L3 autonomous driving is officially implemented, and high-level intelligent driving enters the large-scale application phase. Centered on the technological evolution trend of sensor chips, this report elaborates on the technological direction of autonomous driving sensor chips under three core transformations: perception upgrading driven by regulations, large-scale rollout of L3 from pilot trials, and revaluation cycle of hardware market. It systematically sorts out the technological evolution, market competition and industrial pattern of five major categories of sensor chips including automotive camera, LiDAR, radar, ultrasonic radar and UWB near-field sensing, covering product updates of major suppliers and application status of OEMs.
The technological evolution of the five major categories of sensor chips highlights the following:
Camera CIS chips: Evolving from "merely imaging" to "sensing-computing integration" intelligent perception terminals;
LiDAR chips: Entering a new era of "thousands of channels" and 6D full-color perception;
Radar chips: Moving from "edge processing" towards satellite architecture and high-channel count design;
Ultrasonic and UWB chips: Upgrading from "auxiliary perception" to full-scenario collaborative nodes.
Camera Sensor Chips: Resolution Improvement and Intelligent Integration
In 2026, the automotive camera chip market presents three major trends:
CIS is categorized by structural technology into Front-Side Illuminated (FSI), Back-Side Illuminated (BSI) and Stacked structures, and the stacked architecture represents the mainstream direction for the future. Automotive CIS imposes higher requirements on image resolution, dynamic range, low-light imaging capability and multi-camera collaborative processing capability, and needs to meet automotive-grade certification and functional safety standards. In 2025, BSI + 3D stacked structures accounted for over 75% of mass-produced CIS worldwide, with single-pixel size breaking through 0.6 micrometers.
Take SmartSens SC860AT as an example: Built on SmartSens' brand-new CARSens(R)-XR Gen 2 technology platform, SC860AT adopts a single-pixel architecture, integrated with advanced technologies including Lofic HDR(R)2.0 and SFCPixel(R), and supports AB-Exposure(TM) dual-frame exposure control. It has performance advantages such as high frame rate, high sensitivity and wide dynamic range, catering to brand-new image upgrade requirements of automotive front-view, side-view, rear-view cameras, in-cabin occupancy monitoring system (OMS) cameras and E-Mirror electronic rearview mirrors.
2.High Resolution: 8MP Becomes Standard, and 12MP Gathers Pace
In 2026, with the popularization of ADAS, 8-megapixel sensors have become standard configuration for high-level intelligent driving, and 12-megapixel products make a debut.
Take OmniVision OX08D20 as an example: This 8-megapixel CMOS sensor is specially designed for exterior automotive cameras in Advanced Driver Assistance Systems (ADAS) and Autonomous Driving (AD). Via an innovative shooting solution co-developed with Mobileye, it boosts frame rate up to 60 fps, and adopts a-CSP that reduces its size by 50% compared with peer exterior sensors.
3.Integration of HDR and LFM Technologies - 140dB+ Becomes A Mandatory Threshold
For Advanced Driver Assistance Systems (ADAS), high dynamic range (HDR) exceeding 140dB and LED Flicker Mitigation (LFM) have become rigid technical thresholds for automotive CMOS image sensors (CIS). This integrated technology addresses imaging challenges under extreme lighting conditions, ensuring vehicles accurately identify traffic signals and avoid visual misjudgments.
Take SmartSens SC126AT as an example: Developed on the CARSens(R)-XR Plus technology platform, SC126AT features a single-pixel architecture design integrated with SFCPixel(R), PixGain HDR(R) and other advantageous technologies. It boasts high sensitivity, wide dynamic range, low noise and other superior performance metrics, and supports on-chip ISP functions.
LiDAR Sensor Chips: Era of Thousands of Channels and 6D Full-Color Perception
In 2026, the automotive camera chip market witnesses three major trends:
1.Exponential Surge in Channel Count from Hundreds to Thousands
In April 2026, RoboSense launched Phoenix, the world's first native 2160-channel automotive-grade SPAD-SoC; Hesai released Picasso, a 6D full-color chip with up to 4320 channels; Huawei rolled out an 896-channel dual optical path image-level LiDAR. The perception capability of automotive LiDAR has formally transitioned from "point cloud level" to "image level".
Take RoboSense's dual flagship chips as examples:
The Phoenix Chip is the world's first monolithically integrated native 2160-channel automotive-grade SPAD-SoC. It delivers ultra-high resolution of over 4 megapixels and an ultra-long detection range of 600 meters, elevating the perception precision and coverage of LiDAR to an unprecedented level. The chip now has obtained automotive-grade certification and is scheduled for mass production and vehicle integration within 2026, serving as the ultra-high-definition perception foundation for L3 and above high-level intelligent driving.
The Peacock Chip is the industry's first production-ready all-solid-state large-area SPAD-SoC with a resolution of 640X480. With an ultra-wide field of view of 180°X135° and millimeter-level accuracy, it enables short-range, blind-spot-free point, ultra high-density point cloud coverage, offering unprecedented cost-effectiveness. Scheduled for mass production in Q3 2026, it will be mainly oriented to automotive side and blind-spot detection LiDAR, robotics and spatial intelligence markets to drive large-scale adoption of high-performance perception.
2.Digital Architecture Fully Replaces Analog Architecture
From 2024 to 2025, the automotive LiDAR industry underwent a generational transformation of core architectures. Digital architectures represented by SPAD-SoC (Single-Photon Avalanche Diode - System-on-Chip) have fully replaced traditional analog architectures and become the standard solution for high-performance automotive LiDAR.
SPAD-SoC adopts a sensing-computing integrated architectural design, integrating SPAD arrays and a complete digital signal processing system on a single chip. SPAD-SoC can complete the full workflow including photon detection, time measurement, signal processing and data output simultaneously, directly outputting depth/image data upon photon incidence.
Compared with traditional discrete solutions, SPAD-SoC features high integration, low power consumption, compact size and fast response speed. It acts as the core component for long-range high-channel LiDAR and short-range Flash LiDAR, exerting the most significant impact on LiDAR performance, thus evolving into a critical technological highland contested by all LiDAR suppliers.
3.Chip-Based 6D Full-Color Perception
In 2026, chip-based 6D full-color perception marks LiDAR evolves from a "color-blind measuring instrument" to a "full-color observer". It upgrades from merely outputting three-dimensional spatial coordinates (XYZ) to six-dimensional perception capability that synchronously outputs spatial coordinates and color information (RGB).
At the chip level, color light sensing (RGB) and TOF ranging (XYZ) achieve native pixel-level fusion, directly generating colored point clouds. This means every data point inherently carries precise spatial position and authentic color information upon generation - no stitching, calibration or supplementary deduction required.
Take Hesai's Picasso 6D full-color LiDAR chip as an example: Hesai embeds RGB color light sensing functionality directly into the receiving chip of LiDAR. The Picasso SPAD-SoC adopts an all-fusion design where a single chip handles both near-infrared laser ranging and visible light sensing. Each pixel unit can both emit lasers to measure distance and capture light to identify colors. This means that when laser beams reflect off objects, the chip instantly captures both the spatial position and color of each point.
Radar Sensor Chips: Satellite Architecture and High-Channel Design
In 2026, Satellite Architecture emerges as the core technological transformation direction for automotive radar sensor chips. This architecture subverts the traditional "edge intelligence" design philosophy, driving radars to evolve from independent black box to open perception frontend.
The essence of satellite architecture is a "division of labor model" for intelligent driving perception systems. Radars act as "satellites" dedicated to frontend perception (RF transceiving and raw data collection), while complex signal processing, target detection and tracking algorithms are entirely migrated upward to the central domain controller.
There are two primary satellite radar solutions:
Dedicated RSP IC Solution: Adopts dedicated Radar Bridge chips featuring high energy efficiency and flexibility, led by radar module suppliers;
RSP IP Integrated ADAS SoC Solution: Integrates radar signal processor IP into ADAS SoC for maximum integration, led by OEMs.
Take Infineon RASIC(TM) CTRX8188F as an example: This 8T8R automotive radar transceiver targets imaging radar and high-level ADAS applications. Such high-channel radar chips apply not only to conventional radar modules but also central computing architectures, transmitting rich raw radar data to backend computing platforms.
The CTRX8188F delivers multiple industry-leading RF performance metrics: transmitting power of 14.5dBm ensuring long-distance radar detection; phase noise as low as -100 dBc/Hz for superior signal-to-noise ratio; noise figure of 10.2dB delivering industry-leading receiving sensitivity; 4GHz ultra wideband enabling centimeter-level distance resolution; modulation speed up to 200 MHz/µs meeting detection requirements for high-speed moving targets.
Ultrasonic and UWB Near Field Perception Chips: Upgraded from Auxiliary to Core Perception
Integration of Edge AI Inference Capability: In 2026, mainstream ultrasonic radar modules integrate microcontrollers equipped with Neural Processing Units (NPU), enough to support real-time inference of lightweight deep learning models.
Cross-Sensor Modal Alignment and Fusion: The collaborative capability of ultrasonic radar chips evolves from inter-sensor coordination to multi-modal fusion. In high-end intelligent driving architectures in 2026, short-range high-precision distance data from ultrasonic radars, rich texture information from cameras and velocity vector data from radars are fused via Transformer architecture with cross-modal attention mechanisms.
With the combination of millimeter-wave and UWB technologies, UWB technology is striving to deliver "full-scenario safety protection" for vehicle usage, covering from car search, unlocking, and obstacle avoidance in ADAS operation to parking, door opening and locking. This represents not merely superposition of functions, but a reconstruction of system efficiency and cost structure. UWB is evolving from a single "smart key" to an "intelligent neural network" connecting internal and external perception networks.
Evolution from "Communication Chips" to Communication-perception Integrated SoC
The "three-in-one" integration of chip architecture means that mainstream UWB chips have evolved from communication chips only supporting ranging to a "positioning + radar + communication" trinity SoC solution.
Evolution from "Single-Point Perception" to Collaborative Perception
Traditional UWB radar anchor points operate independently, creating near-field blind spots. By adopting the Sensing technology defined under IEEE 802.15.4ab standard, Calterah realizes multi-static radar mode - rear anchor R1 transmits while R2 receives and vice versa, transforming isolated UWB radars into networked radars operating in coordination.
Take Calterah's UWB product series as an example: Calterah launched two mature development kits: one for in-cabin Child Presence Detection (CPD), and the other for parking assist systems based on 4ab Sensing technology. These kits deliver a complete set of hardware reference design, software SDK, AI model and toolchain to Tier 1 suppliers and OEMs.
The development kits integrate multi-functional fusion solutions including digital key, kick-to-open tailgate, in-vehicle child detection, external parking assist and sentry mode to form an engineering solution ready for rapid deployment. By reusing a set of vehicle-body UWB anchor points, multiple intelligent perception and connectivity scenarios can be enabled.