What is a neuromorphic vision sensor?

The neuromorphic vision sensor is a camera based on the principle of bionics, which can capture key information in a scene and reduce data redundancy and delay. This event-driven sensor brings autonomy closer to reality, and has found its place in high-speed vision applications, such as industrial automation, consumer electronics, and autonomous vehicles.

"Why is it that event-driven visual sensors are neuromorphic?" said Pierre Cambou, chief analyst at Yole Développement (Lyon, France). "Since each pixel is a neuron, it makes sense to introduce AI into pixel processing."

The neuromorphic vision sensor industry, which has been silent for many years, has begun to make a comeback in recent months. Last November, Samsung filed a trademark application for its Dynamic Vision Sensor (DVS) technology for mobile devices and tablet applications. Cambou believes: "This is a bit unexpected, because Samsung's development of DVS technology is mainly for automotive advanced driving assistance systems."

In December last year, Sony quietly acquired the Zurich-based Insightness company. The company's vision sensor can detect motion within milliseconds, even if the sensor itself is in a moving scene. In February of this year, after raising an additional $28 million, Paris-based Prophesee announced at the International Solid-State Circuits Conference (ISSCC) that it would jointly develop a new type of sensor with Sony, the event-driven stacked vision sensor.

Neural mimicry induction originated from the development of the "silicon retina", which mimics the human retina. It was originally proposed in 1991 by Misha Mahowald of ETH Zurich and the Institute of Neuroinformatics. Mahowald explained, “This type of silicon retina subtracts the average intensity level from the image and reports only the temporal and spatial changes, thereby saving bandwidth.” This inspiration inspired the concept of dynamic vision sensors (DVS) and has driven countless startups in recent years. The company is involved. Switzerland's iniVation company is one of them.

Zurich-based iniVaTIon was founded in 2015 by a pioneer in the field of event-driven vision. It has now developed a dynamic vision platform that integrates hardware and software for high-performance machine vision systems. Its neuromorphic DVS chip (model DAVIS346) imitates the characteristics of the human retina, and only when local pixel-level changes occur, data is transmitted, thereby achieving a microsecond-level time-accuracy event stream, which is equivalent to a traditional visual sensor, but The amount of data is much less. The company claims that due to the concept of local processing, the chip greatly reduces power consumption (up to 90% reduction), data storage and computing power requirements, while also increasing the dynamic range of the sensor (over 120 dB).

iniVaTIon has established a network of more than 300 customers and has collaborated with researchers from the University of Pennsylvania, the University of Zurich, and the Defense Advanced Research Projects Agency on the IBM TrueNorth brain-like chip. This research is dedicated to achieving autonomous flight of UAVs. A plan of the European Union focuses on sustainable smart city projects.

Envision a smart factory

At the same time, a silent revolution is taking place in the factory. Autonomy and automation go hand in hand, and machine vision is the supporting force behind the continuous development of manufacturing automation. Unlike simple sensors, machine vision sensors generate large amounts of data to identify defective systems, understand their defects, and intervene quickly. The result is cost savings and increased productivity.

iniVaTIon claims that its dynamic vision platform is very suitable for industrial vision. It can perform high-speed 3D infrastructure scanning for predictive maintenance, high-speed production inspection, particle detection, and microscopes for fluorescence imaging and human motion analysis. In other words, it can perform ordinary or complex repetitive tasks with high speed, high precision and high consistency.

Kynan Eng, CEO of iniVaTIon, said in an interview: "It took us a lot of time to develop a suitable strategy." While other companies are committed to high-speed counting, Eng believes that "high-speed counting is not difficult." Because traditional cameras can achieve "capture a thousand frames or more" per second. If an application does not need to respond immediately, then "use our sensors will be meaningless."

What matters is latency, not data throughput, "our sensors have fast response times." For example, "If you have a robot that needs to move constantly to complete a task, it must adjust its route in real time. The faster it adjusts, the faster it can move and the sooner it detects its own errors."

"I think industrial vision is a relatively low-risk market but a small market," Eng said. Therefore, venture capital institutions are not interested in this. But iniVation sees the potential for organic growth and thinks from the perspective of economies of scale. Through the partnership established with Samsung in 2019, iniVation has shifted from manufacturing and selling chips to providing cameras for the machine vision industry. Yole analyst Cambou pointed out, "You can sell a chip for $100, or you can package the chip in a camera and sell a camera for $1,000."

By developing towards the system, iniVation is improving its position in the value chain.

"We realized that being a chip company didn't make any sense to us," Eng said. "We can raise a billion dollars in funding, but it is still not enough to develop our own chips. Some people want to know why our cameras are so expensive and how can we make them cheaper?" Through cooperation with Samsung, we "solved this." problem."

Increasing quality requirements have accelerated the development of machine vision in the food, packaging, consumer electronics, aerospace and automotive industries. Eng said that iniVation’s goal is to enter a larger market.

If you want to know more, our website has product specifications neuromorphic vision sensor, you can go to ALLICDATA ELECTRONICS LIMITED to get more information