​The MLSoC™ Modalix provided by SiMa.ai is a second-generation SoC (System-on-Chip) designed for next-generation embedded AI applications. While the first-generation MLSoC™ was specialized for inference of convolutional neural network (CNN) models, Modalix is a high-performance SoC manufactured using a 6nm process​ ​that enables generative AI to operate with low power consumption of less than 10W.

In addition to the ability to process generative AI such as large-scale language models (LLM) and multi-modal models (LMM) in an integrated manner, the inference performance of CNN models has also been improved, resulting in an expected performance improvement of approximately 2x compared to the first-generation MLSoC™.

Key components include an 8-core Arm Cortex-A65 (1.4GHz), a Synopsys ARC EV74 vision processor, an Arm Mali ISP, an AV1/H.264/H.265 compatible video codec, PCIe Gen5 (8 lanes), MIPI CSI-2 (4x4 configuration), and 10GbE x4, providing high performance and flexible I/O configurations.