Skip to Content

Edge AI Chips

WQ5008

New multi-modal 3D vision processing chip

The chip is currently the only one in the industry that can simultaneously support binocular ToF, binocular structured light, structured light +ToF fusion upgrade solution, created variety of 3D vision core operator hardening acceleration processing, with high throughput, strong low-power computing power and leading high integration; Self-developed multi-channel data acquisition and neural network depth algorithm, support in complex environment to achieve high frequency sensing computing, to the greatest extent to meet the customer's more application needs and innovation direction.

Features:

  • Built-in self-developed high-performance RISC-V multicore architecture, support enhanced computational capability upgrades and SIMD ISA improvements.
  • Newly self-developed ISP algo and image preprocessing modules, support image processing and arbitrary angle rotation, affine transformation, scaling, and distortion correction of images.
  • Further improved 3D vision hardware processing module, in addition to structured light, added TOF deep processing module, and binocular processing operator.
  • Support 3-channel 1080P@30fps high-resolution image acquisition, high-performance image processing, and H.264/MJPEG encoding at 1080p@30fps.
  • Support financial-grade security encryption engine with complete TrustZone solution, effectively preventing various side-channel attacks, providing customers with all-round data asset protection.
  • Ultra-low power consumption, dynamic power management, with power consumption metrics leading the industry, making it highly cost-effective for battery applications.

WQ5007

Leading-edge chip supporting 3D facial recognition

WQ5007 is the only edge-side and low-power AI computing platform in the AI industry that supports all mainstream 3D vision solutions and is capable of high-speed convolutional neural network computing. It supports live detection locally based on 3D structured light and image recognition based on neural networks, featuring significant advantages such as ultra-low power consumption, balanced computing power, high integration, and fast startup. WQ5007 has dominant advantages in 3D facial recognition and other lightweight IoT applications, perfectly meeting customers' demanding requirements on low-power computing.

Features:

  • Supports live detection locally based on 3D structured light and image recognition based on neural networks.
  • Adopts ultra-low-power design, with the dynamic power consumption less than 400 mW and the standby power consumption less than 100 uW.
  • Flexibly outputs offset, depth, and 3D point cloud.
  • Supports MIPI Tx high-speed output to external AP systems.
  • Outstanding computing power: INT8 200 GOPS, INT16 100 GOPS.
  • Fast startup: Fast boot time < 100 ms.
  • High-level integration: The 3D structured light processor and 32 MB DDR memory are supported.

WQ5301

Car-grade speech recognition chip

WQ5301 supports functions including local voice wakeup, acoustic source localization, directional recognition, offline speech recognition (pure offline dictation), offline natural language processing, offline always on (based on offline NLP), arbitration between offline and online results, offline voice broadcasting, and active offline interaction.

Features:

  • Built-in four 32-bit RV32IMAFC CPU @400 MHz.
  • Adopts the dedicated NNU processor, supporting INT8 200 GOPS.
  • Supports DDR3/DDR3L@1066 Mbps and the capacity can be expanded to a maximum of 1 GB.
  • Provides various acoustic interfaces, and supports analog MIC, I2S, and TDM functioning as acoustic input.
  • Supports eMMC-based local firmware boot and USB-based remote firmware boot.