Emza and Alif collaborate for AI vision in low power devices
Complex artificial intelligence (AI) inference capabilities such as eye tracking and facial identification in low power, low cost devices has been demonstrated by Emza Visual Sense and Alif Semiconductor.
The two companies have collaborated on Arm-based hardware and optimised models to make AI at the edge a reality. The companies are demonstrating Emza’s trained face detection model running on Alif’s Ensemble microcontroller, which is the first to feature the Arm Ethos-U55 microNPU. The Emza model runs an order of magnitude faster on the Ensemble device with Ethos-U55 compared to a CPU-only design, said Alif.
A face detection neural network model (a light version of the single shot detector architecture) extracts a face bounding box and face yaw angle. It executes on the Alif silicon with Ethos U55 core in four milliseconds, compared with 394 milliseconds executing on silicon with an M55 CPU only (at 400 MHz). On the high efficiency, power-optimised Ethos U55 core, the model executes in 11 milliseconds.
The combination of AI/ML (artificial intelligence / machine learning) frameworks, models, neural networking processors (NPUs) and silicon that are all optimised for edge AI means it is now possible to support complex AI inference capabilities such as eye tracking and facial identification in low power, low cost devices, said the partners. The development creates an opportunity for new applications and use cases across industrial IoT devices, consumer appliances, for example.
“To unleash the potential of endpoint AI, we need to make it easier for IoT developers to access higher performance, less complex development flows and optimised ML models,” said Mohamed Awad, vice president of IoT and Embedded at Arm. “Alif’s [microcontroller] MCU is helping redefine what is possible at the smallest endpoints and Emza’s contribution of optimised models to the Arm AI open source repository will accelerate edge AI development.”
Emza trained a sophisticated, full implementation of a face detection model on the Arm Ethos-U55 microNPU and is the first Arm AI ecosystem partner to contribute a complete application code ML example to Arm’s ML Embedded Eval Kit repository. Companies can use it to gauge run time, CPU demands, memory allocation and other requirements even before silicon is available, confirmed the company.
Commenting on the future potential for TinyML edge devices, Emza’s CEO, Yoram Zylberberg, said: “There is virtually no limit to the types of visual sensing use cases that can be supported by new powerful, highly efficient hardware”.
Reza Kazerounian, president and co-founder, Alif Semiconductor explained: “Ensemble devices leverage one of the latest generation processor cores and neural network accelerators from Arm in highly scalable configurations. On top of that, we layer Alif’s unique technologies, low power design techniques, deep embedded security, and a high level of functional integration – all resulting in extremely capable, secure devices with low power consumption and long battery life”. The collaboration showcases the capabilities of devices and empower the next generation of edge AI applications.
The full face detection project example for the Arm Ethos-U55 is available now and other Emza object detection models are available on GitHub.
The Arm Ethos-U55 microNPU is available now and Alif Semiconductor is sampling the Ensemble family now.
Emza Visual Sense specialises in low power edge AI devices. The company provides hardware, software, algorithms and IP to semiconductor companies and OEMs to bring AI capability to tiny, power- and cost-constrained edge devices. As compute power increases and silicon costs decline, the company believes the market for these tiny edge AI devices is rapidly expanding across a broad array of segments such as consumer, industrial, automotive and smart cities.
Emza’s WiseEye low power vision AI systems combine its computer vision and ML algorithms, CMOS imaging sensor and tiny AI SoC. They are shipping today in popular consumer products.
Emza is wholly owned by Himax Technologies.
