AUTOMOTIVE
issue up to eight tasks to the processing unit for fully secure multitasking, ensuring the key processes performance is being prioritized. In the more distant future, when fully autonomous driving becomes widespread, cars could start integrating Metaverse- like solutions in their cockpits. The first implementations will likely focus on changing the in-vehicle environment, with passengers being able to interact with the Metaverse either through bigger more immersive displays or even VR headsets as they travel to their destination. This would make powerful GPU silicon even more of a requirement along with cloud-enabling solutions.
Intelligent and autonomous As efficient electric motors are replacing combustion engines, central processing is starting to remove human error from the driving experience, while enabling safe ADAS and AD solutions. The Society of Automotive Engineers defines six levels of automation – from Level 0, which has no automation systems at all, to Level 5, where the vehicle performs all driving tasks in all conditions and a steering wheel is not even required. The path to Level 5, albeit long, will require a combination of advanced AI, leading-edge software and high-performance compute to help process massive amounts of data from multiple sensors including cameras, lidar and radar. Beyond autonomy, AI will play an increasingly important role in most aspects of the car – from in-cabin monitoring that assesses driver attentiveness and fatigue to language processing for voice commands. AI systems can even scan the road ahead, adjusting vehicle suspension to accommodate the driving experience accordingly. However, ADAS and AD remain the most challenging tasks for AI, as these solutions are data processing intensive. The compute
requirements for Level 4 autonomy require hundreds of TOPS, while Level 5 autonomous driving needs 1000 TOPS worth of performance. Dedicated silicon, such as Imagination’s neural network accelerator (NNA) family, can deliver this in a power and area-efficient way. Leveraging highly scalable multi-core, multi- cluster architectures, OEMs can use solutions such as the IMG Series 4 NNA to harness performances greater than 2000 TOPS; a platform enabling the real-world deployment of autonomous driving systems. The key benefit is that the NNAs can use on-chip memory and shared SRAM alongside advanced scheduling techniques to enable multiple neural networks to be run concurrently, offering incredible performance at low latency and limited system bandwidth requirements – vital for quick response times in autonomous cars.
Lightning-fast data management Just one camera can require more than 1Gbps of data transfer speed, and many modern ADAS-equipped vehicles feature up to 16 cameras in the front of the vehicle and bumper alone. When combined with the slew of other potential sensors found in modern vehicles, this results in terabytes of data needing to be managed across a complex system. To achieve this, manufacturers need to look beyond CAN bus – which has been in use since 1987 and provides bespoke but low-bandwidth designs – as this technology cannot keep up.
To manage data of this magnitude, OEMs need to look at other well-established protocols – Ethernet. Just a twisted pair of modern IEEE standard Ethernet cables can move up to 1Tbps around the vehicle. Thanks to its open standard and universal availability, Ethernet can reduce the cost of in-vehicle connectivity by up to 80%.
Cabling is just half of the equation. Creating reliable network infrastructures to deliver data to CPUs is critical for the safety and integrity of the system. To achieve this, OEMs need a switch and router, which can handle data throughput and provide effective distribution. Imagination’s Ethernet Packet Processor (EPP) is a great example of how this can be achieved in low-power and low-area conditions while maintaining high- performance and minimum CPU loads. With 12 years of history behind, it has been deployed successfully in ASIL-D silicon in the automotive market, following ISO 26262 processes.
Unlocking the future with silicon The underlying architecture of the modern car is changing. Three areas will remain key in driving successful automotive innovation for electronic vehicles – HMI, AI and data management. Human-machine interfacing is evolving at a rapid pace, with newer GPUs unlocking advanced processing and ray tracing, while AI remains the foundation for functionally safe assisted or autonomous- driving features – with Ethernet underpinning all the required data transfer.
With over 45% of the in-car graphics market share owned by Imagination, it’s clear that silicon is of tremendous value for future automotive developments. Manufacturers are now looking to differentiate offerings through HMI, all while being at the leading edge of AI features – including ADAS and AD. The resulting push for innovation in GPU, NNA and EPP technologies will have an impact across industries.
It’s an exciting time for electric vehicles and the silicon industry especially, as more OEMs are taking advantage of the lower barriers of entry for EV manufacturing, shaping the future of automotive. As Level 3 autonomous driving regulations kick in, the era of fully self- driving cars is ever closer and bringing thrilling technological developments with it.
Imagination Technologies
www.imaginationtech.com
MARCH 2022 | ELECTRONICS TODAY 9
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46
