If only it had a Brain
Artificial Intelligence at the edge
From predictive analytics to autonomous navigation and many applications in between, advances in AI technologies have enabled and benefited applications across all sectors. Deep learning, a subset of Machine Learning, has been the workhorse in this space but the models are compute and resource intensive. Traditionally, the training of AI algorithms would take place on server farms in the cloud with huge datasets.
The sheer size of some of these models and the compute required (usually in Tera OPS) had relegated inference to the cloud as well, even when concerns over latency, privacy, security and bandwidth abounded. Over the last couple of years, advances in both algorithms and in edge hardware has enabled some of these inference engines to be run on low resource end devices enabling edge AI.
A typical AI device comprises of a sensor or sensors, a connectivity option, a compute element and power management. While a lot of attention is focused on the compute element that runs algorithms or appropriately packages the data for consumption by upstream AI algorithms, the other components have a significant bearing on the quality of the actionable inferences that these devices are designed to generate.
ON Semiconductor’s family of sensors, signal processors and connectivity SoCs are designed to help our customers realize robust and differentiated AI devices. A good example of this is the Image sensor family from ON Semiconductor. With built in features such as High Dynamic Range (HDR), the sensors allow the operation of AI cameras in a variety of lighting conditions. The kinds of lighting and other environmental conditions encountered by autonomous vehicles requires technologies such as HDR, Global shutter and in some cases, a single sensor capable of RGB and IR imaging.
Our SiPM (Silicon Photo Multipliers) sensors have the industry leading sensitivity and allow for building of robust LiDAR devices. Outside of automotive, LiDAR is being adopted in robots and other industrial applications that require accurate depth and/or obstacle avoidance.
The audio processor family of products provide a great example of highly integrated, small footprint and highly efficient edge AI devices. With the ability to recognize words and phrases, many automation use cases encountered within smart homes and buildings can be effectively run on these devices locally. These full featured devices can also connect to the cloud where true Natural Language Processing can be run.
A key figure of merit for edge AI devices, especially if they are battery operated, is their efficiency or power consumption that directly translates to battery life. Most of our products, including image sensors and audio processors are the most efficient in the industry. These products nicely complement the ultra-low power wireless PAN technologies from ON Semiconductor and help our customers design truly efficient and high performance edge AI devices.
Complementing a broad sensor portfolio is the strong ecosystem support. The ecosystem partners we work with include some of the most recognized names in the industry and through this collaboration we enable significantly reduced development times for customers designing and deploying edge AI devices.
Automation, machine vision, predictive maintenance, and other Industry 4.0 applications are becoming pervasive across industries big and small, resulting in increased operational efficiencies, productivity gains and cost savings. Due to privacy, security and latency concerns, the AI and ML algorithms underpinning the explosive growth in these applications are typically run on edge devices. High performance sensors are critical for both training and inference of AI algorithms. ON Semiconductor’s image sensor family is ideal for demanding machine vision applications and features:
- Global Shutter sensors that eliminate motion artifacts – critical for embedded vision applications
- High Dynamic Range (HDR) image sensors that allow operation in demanding real life lighting conditions
- RGB + IR sensors and single sensors that can estimate depth
- High gain, excellent Photon Detection Efficiency (PDE) and fast timing and high sensitivity SiPM (Silicon Photo Multipliers) for LiDAR applications
ON Semiconductor’s family of audio processors are extremely energy efficient and incorporate enough compute power to run AI based Voice User Interface (VUI) on the edge. These full features devices feature multiple cores and are supported by a comprehensive software suite.
Transportation sector is on the cusp of a major upheaval thanks to AI fueled innovation in autonomous navigation. Robots navigating complex environments, self-driving cars and trucks drastically cutting down accidents, collision avoidance and subject tracking drones are all becoming a reality. Due to latency and bandwidth concerns (and security), the trained AI models have to run on edge devices. Advances in edge hardware, algorithms and sensors that perform even in the harshest of ambient conditions are key to enabling these devices. ON Semiconductor’s family of sensors with industry leading performance and cutting edge technologies ensure robust edge devices. These include the following:
- High Dynamic Range (HDR) image sensors that allow operation in demanding real life lighting conditions
- Global Shutter sensors that eliminate motion artifacts
- Ultra-low power sensors for battery operated edge AI devices
- RGB + IR sensors and single sensors that can estimate depth
- High gain, excellent Photon Detection Efficiency (PDE) and fast timing and high sensitivity SiPM (Silicon Photo Multipliers) for LiDAR applications
ON Semiconductor’s products and technologies coupled with a strong ecosystem of partners that we engage with result in robust and differentiated edge AI devices and a reduced time-to-market.
