Summary of I.MX8M MINI BASED MODULE FEATURES GRYFALCON NEURAL ACCELERATOR
The i.MX8M Mini SOM is a compact System-on-Module designed for Edge AI and IoT applications. It integrates NXP's Arm Cortex A53 processor with Gyrfalcon's Lightspeeur 2803S Neural Accelerator, supporting TensorFlow, Caffe, and PyTorch frameworks. The module features up to 4GB LPDDR4, Bluetooth, WiFi, PCIe 2.0, and robust multimedia capabilities including 1080p video encoding and decoding, making it ideal for high-performance, low-power embedded systems.
Parts used in the i.MX8M Mini SOM:
- NXP i.MX8M Mini SoC
- Arm Cortex A53 processor
- Cortex M4 general purpose processor
- Gyrfalcon Lightspeeur 2803S Neural Accelerator
- LPDDR4 memory (up to 4GB)
- USB 2.0 ports
- Bluetooth connectivity
- Optional WiFi module
- PCIe 2.0 interface
- MIPI-DSI display interface
The i.MX8M Mini SOM – Building Block with Embedded Artificial Intelligence Capabilities
i.MX8M Mini SOMs offer a compact System-on-Module platform with robust processing power and artificial intelligence acceleration using Gyrfalcon’s Lightspeeur® 2803S Neural Accelerator – designed for next generation Edge AI applications using the standard TensorFlow, Caffe and PyTorch model development frameworks.
SolidRun’s i.MX8M Mini SOMs harness NXP’s Arm Cortex A53 single/dual/quad core 1.8Ghz (with single Cortex M4 general purpose processor), i.MX8M Mini SoC built with advanced 14LPC FinFET process technology. This cutting-edge building block is tailor made for a wide range of IoT and industrial applications, featuring up to 4GB LPDDR4, 2 x USB 2.0, powerful network connectivity options including Bluetooth and optional WiFi, PCIe 2.0 and robust multimedia features including 20 audio channels (32bits), MIPI-DSI, and 1080p encoder and decoder.
Numbers referenced below are for Gyrfalcon 2803S chip.
- High performance : 24 TOPs/Watt
- Up to 16.8 TOPs @ 300MHz
- Low Power: 16.8 TOPs @ 700mW
- Support for: ResNet, MobilNet and ShiftNet
- Supported standard AI frameworks such as TensorFlow, Caffe, PyTorch including pre-trained AI models
I.MX8M Mini Advantages
Processing Power
The i.MX8M Mini SOM based on NXP’s robust i.MX8M Mini Arm Cortex A53 processor offers a range of features tailor made for embedded Artificial Intelligence and IoT. With AI Acceleration and a range of connectivity features – SOM i.MX8M Mini is perfect for Edge AI, IoT and next generation artificial intelligence applications.
AI Acceleration
i.MX8M Mini SOM is ready to fuel the next generation of Artificial Intelligence applications – harnessing a powerful deep learning CNN AI acceleration microprocessor. The 9 x 9mm accelerator, based on Matrix Processing Engine architecture, offers multi-dimensional processing for extremely high speeds at very low power (24 TOPs/Watt).
Read more: I.MX8M MINI BASED MODULE FEATURES GRYFALCON NEURAL ACCELERATOR
- What is the primary purpose of the i.MX8M Mini SOM?
It is designed for next generation Edge AI applications using standard TensorFlow, Caffe, and PyTorch model development frameworks. - How much processing power does the AI accelerator offer?
The accelerator provides up to 16.8 TOPs at 300MHz and achieves 24 TOPs per Watt efficiency. - Which AI frameworks are supported by this platform?
It supports standard AI frameworks such as TensorFlow, Caffe, and PyTorch including pre-trained AI models. - Can the i.MX8M Mini SOM handle video processing tasks?
Yes, it features powerful multimedia options including a 1080p encoder and decoder. - What type of processor cores are included in the SoC?
The SoC includes single/dual/quad core Arm Cortex A53 processors running at 1.8Ghz plus a single Cortex M4 general purpose processor. - Does the module support wireless connectivity options?
Yes, it offers powerful network connectivity options including Bluetooth and optional WiFi. - What is the maximum amount of LPDDR4 memory available?
The system features up to 4GB of LPDDR4 memory. - Which neural network architectures are supported by the accelerator?
The chip supports ResNet, MobilNet, and ShiftNet architectures.
