1、High-performance and energy-efficient DynamAI NN engine (used in multi-camera access and high-performance computing scenarios):
Hybrid-precision 4-bit/8-bit MAC array
Average 80% MAC array utilization rate for convolutional layers
Sparse support for storage and acceleration performance
Fixed-point 8/16-bit and floating-point 16-bit GEMM and non-linear function acceleration
Open and family-oriented DynamAI NN compiler tools
2、Enhanced version of automotive-grade NeurallQ ISP pipeline:
12-channel HD camera access
Support for cameras with resolution exceeding 17MP
3-exposure HDR combination, dynamic range up to 140dB
Low-light offline noise reduction and LED flicker suppression
Added RGB-IR Sensor support
Process 2.4G pixels per second
3、Information and Functional Safety:
The chip complies with ISO 26262 ASIL-D and AEC Q100-Grade 2 standards
Both CPU and GPU support dual-core lockstep
Real-time safety monitoring and detection
Built-in safety protection measures such as ECC (Error-Correcting Code) and Parity
EVITA-Full certification and China National Cryptography Second-Level certification.
Secure boot
On-chip OTP (One-Time Programmable memory) for key storage and lifecycle management
Multiple independent hardware encryption/decryption engines and dedicated hardware acceleration engines for secure communication.