WTM2101 CIM SoC Chip
-
-
Optimized for Energy-Efficient Scenarios (Battery/Small Devices)
KK-Level Mass Production
- >50Gops Computing Power
- 5uA-3mA Ultra-Low Power Consumption
- 1.8M Max Parameter Capacity
- 2.6mmx3.2mm Size
Application scenarios
100+ Words Speech Recognition（offline）
Computing Power：200M-1Gops
Supports ASR Model
Supports continuous recognition of 40 to 300 words
Supports ASR Model
Supports continuous recognition of 40 to 300 words
AI Noise Reduction
Computing Power：1Gops
Power consumption：1.5mA
Optimal voice retention + intelligent noise reduction
Power consumption：1.5mA
Optimal voice retention + intelligent noise reduction
Environmental Sound Recognition
Low-Power Multi-Classification
Real-time analysis of ambient sounds
Automatically switch configuration parameters
Real-time analysis of ambient sounds
Automatically switch configuration parameters
Keyword Wake-Up（offline）
Computing power:10-20Mops
Wake Word (KWS) model
Generally supports 1 to 20 words
Wake Word (KWS) model
Generally supports 1 to 20 words
AI Transparency Mode（offline）
Low-latency computing scheme（<5ms）
Adaptive noise reduction
Effectively suppress noise and provide transparent and effective information
Adaptive noise reduction
Effectively suppress noise and provide transparent and effective information
Anti-Feedback Suppression
Ultra-low power consumption enhances howling suppression
Traditional algorithm +NN algorithm
Howling can still be prevented at tens of dB gain
Traditional algorithm +NN algorithm
Howling can still be prevented at tens of dB gain
WTM2101 Test Board
Test the CIM matrix operation (maximum matrix operation scale 896x1024)
Test the programming and operation accuracy of CIM (8-bit)
Test the operational efficiency of CIM
Test the operation effect of deep learning networks (including mainstream deep learning operation operators)
Complete the demo demonstration: command word recognition, voice noise reduction, voiceprint recognition, etc
