NEUCHIPS Coporation 中文

Detailed solution introduction

NEUCHIPS digs into the core architecture design of deep learning algorithm and architecture design. The team has a deep competitive advantage of algorithm-architecture-hardware-software co-optimizing ability.

NEUCHIPS solutions can accelerate AI voice, vibration, video, and data processing:
1. RNNAccel: RNN accelerator with high accuracy compression engine
An hardware accelerator IP for deep learning model computing. It support LSTM, GRU, Fully Connect, MLP, and Vanilla RNN. Including a patent compression technique which can huge reduce memory footprint. By RNNAccel develop tools, the ASIC vendors can quick integrate NN models, easy deploy configurations, and verify designs. RNNAccel is extremally suitable for voice, vibration, and sequential data analysis such as ANC, ECG monitoring and symptom detection, keyword spotting. It is perfect for a class of smart, battery-powered, always-on ASICs such as TWS/ANC earphones, smart speakers, voice assistance, wearable devices, voice translation. NEUCHIPS provides RNNAccel IP licensing and customize integration services.

2. HarDNet: World's fast real-time sematic segmentation CNN architecture
HarDNet is a fast, low cost, high accuracy, and safe CNN-based neural network architecture which proposed it in ICCV-2019 by NEUCHIPS. It been listed the world's 1st State-Of-The-Art sematic segmentation neural network model in Paper-with-Code. This architecture is also perfect fit for object detection, object classification, and multiple object tracking applications. NEUCHIPS provides HarDNet algorithm, software, IP licensing and consulting services.

3. RecAccel: World's first DLRM DSA
RecAccel is a DSA based on Facebook STOA recommendation solution - Deep Learning Recommendation Model (DLRM). For eCommerce, on-line advertising, media recommendation applications, it can provides fast and efficient recommendation to target users. We provide RecAccel FPGA accelerating board, IP licensing, and customized integration services.