Energy-Efficient Design Helps Autonomous Vehicles Take Off

A look at the impact of power, thermal and structural integrity challenges.

popularity

A new flight endurance record was set recently by the Predator B/MQ-9 Reaper Big Wing experimental drone, traveling 37 hours non-stop, and setting the bar for the next stage of autonomous flight development. The growth of drone technology and their autonomous capabilities for surveillance, security, and reconnaissance has grown rapidly over the past several years, thanks to ever-increasing performance and scalability of electronic systems. However, the dense concentration of high-performance processors, antennas, and sensors create a growing challenge for power efficiency and thermal management for these autonomous systems, which need to perform reliably in rugged environments.

As the need for more advanced autonomous capability has grown, more compute power is needed to support communication, high-resolution images, data collection, and deep learning to help drones navigate and respond to various situations and environmental conditions. Powerful compute processors and specialized hardware accelerators are required to enable the necessary performance. However, this compute power comes at the price of increased thermal cost.

Energy efficiency and thermal management are the key enablers for these systems. The size (and weight) limitations of UAVs demand an integration of compute systems in tight spaces. It places groups of specialized convolution engines like GPGPUs, and high-performance CPUs, in close proximity with limited space for extensive cooling solutions. Any successful thermal strategy will need to account for cooling these power-hungry components during flight as well as during take-off and landing from desert bases.

Power consumption directly impacts the total heat dissipated by the system, and therefore is a critical concern as engineers make design tradeoffs to optimize the cooling solutions for their target system. Any cooling solution (i.e. air-cooled vs. liquid-cooling) will impact the size, weight, and cost, and ultimately the system performance. Analyzing the thermal impact on electrical reliability enables designers to select the appropriate chip, package, and board components, optimize their designs, and improve reliability, while reducing system cost, development time, and the need for multiple design spins.

Likewise, thermal impact on structural integrity is a major concern for vehicle reliability, requiring failure analysis for thermal stress and deformation on the board, as well as structural integrity concerns of thermal cycling. With little margin for error in these autonomous systems, thermal management needs to begin early in the design phase, with a power-efficient design flow, in order to minimize heat dissipation in the system, and support cost/weight trade-offs.

These best practices from UAV system development are now propelling the next generation autonomous vehicle, the driverless car. Similar to UAVs, automotive systems are driven by reliability, cost, performance, and especially safety requirements. Selection of cooling solutions to ensure performance of these complex on-board electronic systems is key to meeting targets for user experience and vehicle reliability, within budget constraints.

Autonomous vehicle technology is rapidly evolving with the introduction of deep learning to handle complex real-world scenarios. As these automotive systems rapidly incorporate more complex driver assist and autonomous capabilities, best practices for ensuring power, thermal, and structural integrity of these systems will continue to fuel rapid innovation of power efficient, small-form factor and high reliability systems to help next-generation autonomous vehicles to take off.



  • Sandeep Patil

    Nice reading Margaret and thanks for the post. One query though about the power optimization in chips. Since ANSYS’ Apache tool is being used widely in the chip design industry for power and rail analysis for some time now, is it possible to reverse engineer the analysis and model know what patterns of a design can cause the failures?