Sondrel, a leading provider of ultra-complex custom chips for leading global technology brands, has optimized its SFA 100 to make it ideal for battery-powered AI at the Edge applications. The SFA 100 is one of the five architectural platforms in the company’s Architecting the Future chip design platform, which provides a ready-made design baseline that the company uses to build a bespoke solution for every customer much faster than starting from scratch each time.
Image Credit: Sondrel
“The drive to AI everywhere was clear at this year’s CES,” said John Chubb, Sondrel’s CEO. “It enables companies to create better, more feature-rich products but augmenting with AI is not a simple task. It requires much more powerful processor cores than a standard version of a device to handle the large amounts of data involved which can’t just be added onto an existing design. It requires a design from scratch to make sure that the right processors, memory, DSPs, NoC, and other components all fit together against the target Power, Performance, Area, and Timing specifications.
“This is where our Architecting the Future approach can save a significant amount of time and engineering effort. We don’t start from scratch every time, as we use existing, proven blocks and layouts. By using one of our pre-defined starting architectures, such as the SFA 100, we can rapidly tailor this to create a unique, custom solution for each customer. Effectively, we are re-using a framework or chassis and scalable entities such as cores that we have used before along with tools, flows, project management, etc., which reduces risk and, most importantly, time to silicon so that the customer can be fast to market with the AI-empowered device.”
The SFA 100 uses an Arm® subsystem with a choice of Arm cores and the option of a security subsystem. Its low power requirement makes it the ideal platform for battery-powered edge AI applications such as smart homes, smart factories, and sensor fusion devices. The device provides the ability to integrate a machine-learning engine supporting a variety of inference-based ‘smarts’ such as voice activation, image classification, gesture recognition, filtering, etc. A radio subsystem, such as a Bluetooth IP block or WiFi, can also be added if required. The customer’s IP and any third-party IP can be included, as Sondrel has the ability to wrap IP blocks so that they have all the necessary interfaces to just drop onto the chassis. More powerful solutions can be created by using the SFA 200 or SFA 300 as starting points.
According to a report by Transparency Market Research, the Edge AI market is predicted to expand at a CAGR of 33.5 % to reach $76 billion by 2031. The key is the drive to process data locally and thus eliminate the latency from transmitting the data to a centralized cloud server for processing. Applications range from Industry 4.0, where it is used to provide real-time decision-making, to devices such as drones, smartphones, cars, etc.