Edge servers, robots and industrial systems now have to run LLMs, VLMs and multi-channel vision on-site. ARIES is Mobilint’s AI accelerator chip, built to deliver 80 TOPS within the power, heat and space limits of real deployments.
Architecture and compiler are designed together, so CNNs, LLMs and multimodal models run efficiently on one chip.
Each layer gets its own precision, keeping accuracy high while using memory efficiently.
Lookup tables handle GELU, SiLU and other modern activation functions at low area and power cost.
SDK qb supports more than 490 validated model architectures, from CNNs, RNNs and LSTMs to LLMs and VLMs, including YOLOv12, Llama and EXAONE.
PyTorch, TensorFlow, ONNX
Hugging Face Transformers API, with 4-bit weight quantization for LLMs
Vehicle detection and license plate OCR at the intersection
Autonomous navigation and vSLAM on board
Defect detection and vision inspection on the line
On-premises LLM and multi-channel video analysis
Local AI development and inference
Voice interaction and visitor analytics
