Christopher Frost explains how scientists are using the ISIS Neutron and Muon Source to simulate the impact of cosmic rays on electric and autonomous vehicles
Cosmic rays—high-energy particles originating from the far reaches of the universe—are constantly impacting the Earth. While these rays pose little risk to humans by the time they reach the surface, they can wreak havoc on the electronic systems in planes, computers, and increasingly, the autonomous and electric vehicles that are reshaping the mobility landscape. They are generated from events like supernova explosions, where immense gravitational forces and extreme temperatures cause atomic nuclei to be stripped of their particles at high energies. These primary cosmic rays then travel towards Earth, colliding with the atmosphere and creating showers of secondary particles capable of disrupting electronics on a microscopic scale.
The effects of these rays start small but can have major consequences. A single cosmic ray particle can cause a computer’s binary code to flip, changing a 0 to a 1 or vice versa. While rare, these “bit-flips” are unpredictable errors that can corrupt data in internet routers, airline instrumentation, and the complex neural networks behind autonomous vehicles. Bit flips occur all the time, but their consequences have been reduced by modern error-checking hardware and software. However, for safety-critical systems like self-driving cars, even the smallest failure probability is unacceptable, requiring more rigorous research into detection and prevention.
Vulnerability in next-generation vehicles
The cosmic ray threat is becoming increasingly pressing as the automotive industry pivots toward electric and autonomous vehicles. These next-generation vehicles offer immense societal benefits but contain vulnerable networks of electronics and sensors. The AI systems behind self-driving cars are highly complex and computationally intensive, and this makes them susceptible to cosmic ray-induced errors. They rely heavily on neural networks for tasks such as object recognition, path planning, and decision-making. Cosmic ray-induced ‘bit-flips’ can corrupt the neural networks, causing them to produce incorrect outputs.
This can lead the car to misunderstand what its sensors are ‘seeing’, with potentially catastrophic consequences. A single corrupted bit could cause the vehicle to misidentify a pedestrian or fail to recognise a stop sign. At best, these errors would undermine public confidence in autonomous technology. At worst, they could lead to accidents or collisions. As automakers strive to bring self-driving cars into the mainstream, they need to rigorously stress-test their systems against the cosmic ray threat.
The AI systems behind self-driving cars are highly complex and computationally intensive, and this makes them susceptible to cosmic ray-induced errors
The same vulnerability applies to the high-voltage electronics and batteries at the heart of electric vehicles. High-voltage, high-power electronic devices are key to the development of electric vehicles, which have seen large growth in the last few years as a result of society trying to reduce carbon emissions. Unfortunately, these devices have been shown to be susceptible to complete ‘burn-out’ as a result of cosmic rays. In these cases, the device completely fails and would need to be replaced. As these power devices are often integrated into larger power systems, one failure can potentially damage the entire system.
One common hardware solution that has reduced the impact of bit flipping is modular redundancy, in which machines are built with multiple components in case a component fails. An electronic system, for example, could be built with three separate processing units so that if one encounters an error, this can be detected and the other two can override the malfunctioning unit. The main issue with this approach is that increasing the number of processing units can greatly impact cost, particularly when dealing with complex electronic systems including those in autonomous vehicles.
The cosmic ray simulator
As cosmic rays strike at random and in the field are almost impossible to trace, they are challenging to identify or avoid. Specialised facilities, like the ChipIr instrument at the ISIS Neutron and Muon Source in Oxford, allows for the simulation of the impact of cosmic rays so that engineers can develop new ways to detect, correct and minimise these errors.
ChipIr is a dedicated machine commissioned in 2017, designed specifically for the purpose of subjecting electronics to ‘cosmic stress-tests’. It is the largest instrument in Europe for this purpose. At the heart of the massive ISIS facility, protons are accelerated to 84% of the speed of light. These are then blasted at heavy metal targets to produce ultra-high-energy neutrons that mimic the specific energies of cosmic rays. ChipIr is able to accelerate the testing of electronics, bombarding silicon microchips with neutrons at 1.5 billion times the natural intensity so that an hour at ChipIr represents over 170,000 years in the real environment.
By developing new detection and correction mechanisms or building redundancy into critical systems, major companies are working to tackle the problem. For example, Infineon is a global market leader in automotive semiconductor solutions, working with products including power semiconductors, microcontrollers, and sensors. Semiconductors are essential for e-mobility, autonomous driving, as well as connectivity and advanced security. Over the past 30 years, Infineon has been working with the ISIS Neutron and Muon Source to ensure the reliability of its devices to meet industry standards, which state that device failure has to occur less than once in a million years. To assess this, thorough testing is required and specialised facilities such as ChipIr can drastically reduce lengthy testing times.
Both companies like Infineon and academic teams are utilising ChipIr, harnessing its powerful neutron beam to understand the reliability of the technologies being deployed in self-driving cars and electric vehicles. This is leading to some innovative ideas to improve reliability, such as injecting faults into AI systems while training the neural networks, to allow the system to familiarise itself with the occurrence of neutron-induced errors. Such ‘fault aware’ training improves the reliability, with virtually no effect on performance. Such insights are enabling researchers to engineer more robust system designs and investigate new ways to enhance resilience against particle-induced failures.
Substantial additional research is essential to guarantee the vehicles of the future are resilient against the cosmic ray hazard. But facilities like ChipIr are paving the way, stress-testing limitations so that automakers can deliver autonomous and electric cars that are safe, reliable and defect-free.
About the author: Dr Christopher Frost is Head of Irradiation at ISIS Neutron and Muon Source
