40 The drive to reduce size, weight and power (SWaP) in autonomous systems is an engineering challenge when faced with greater demand for more complex sensors and the resulting increase in embedded computing requirements. The task of balancing the latter with system and cost specifications covers embedded boards and the selection of the right level of processing power and consumption. The drive to add machine learning to process sensor data is also pushing up compute requirements with computer-vision algorithms based on transformer neural networks. Transformers are a more complex AI framework, requiring higher processing power, and they are being used in centralised control devices for driverless trucks. But this technology is also being adopted in driverless cars, tapping into speech-to-text capabilities as the user interface for passengers. The latest generation of graphics processing units (GPU) are being used as accelerators for these algorithms in the monolithic system-on-a-chip designs that are coming in 2025. New approaches such as chiplets are enabling smaller compute systems with higher performance, providing a mix of capabilities, including different AI accelerators. Such devices for centralised controllers in driverless cars are now emerging as samples, with production scheduled for 2026. But all of these approaches need to address the key issues of reliability and long-term supply. Boards need soldereddown processors and memory to tackle the reliability issues caused by vibration, and the more complex chiplet devices need to be qualified to automotive and aerospace quality standards, such as AECQ-100 and AECQ-200, respectively. Long-term supply often means decades of availability. This can be a key restraint on the adoption of suitable embedded computing technologies for uncrewed systems. Designers are taking new approaches to embedded systems to pack in more power in ever-smaller forms, as Nick Flaherty reports Board behaviour December/January 2025 | Uncrewed Systems Technology New chiplet technologies for autonomous vehicles (Image courtesy of Renesas Electronics)
RkJQdWJsaXNoZXIy MjI2Mzk4