For customers seeking such qualification, Imagination is backing the IP with additional investment “in the people, resources, culture and management structure needed to support automotive solutions,” company marketing director Bryce Johnson told Electronics Weekly. “Our testing, documentation and built-in hardware structure makes it much easier for silicon manufactures to gain ASIL-B accreditation.”
Johnson added that GPUs based on its earlier Rogue architecture are already operating in vehicles, in applications including infotainment, digital cockpit and advanced driver assistance systems (ADAS). Customers have included Denso, Renesas, Socionext and Texas Instruments – In total, according to Imagination, tens-of-millions of devices have been shipped into automotive applications, giving it an automotive application processors market share over 50% by volume.
The new package – hardware IP, software, tools, documentation (including traceability information) and long-term support – has been branded ‘PowerVR Automotive‘.
“It’s designed to ensure customers’ SoCs can achieve the ASIL-levels of functional safety, meeting industry standards such as ISO 26262 and IEC 61506 at the system level,” said the firm.
Its hardware IP is a modification of the firm’s high-end ‘Furian’ architecture and based on its 8XT GPU – which can be implemented with more or fewer processing sub-units depending on how heavy the intended graphics load is.
Called 8XT-A, the automotive core adds three blocks: an error-correcting code (ECC) calculator and ECC-based data corrector – both for memory – and a collection of built-in self-test functions.
“The in-situ tests can run a number of times per second while the processor is running,” said Johnson. “The idea is to find transient errors and ‘stuck-at’ errors in a timely fashion to warn the driver to take the car off the road or, if it is very serious, the car would take itself off the road.”
On top of this, Imagination’s GPUs can be implemented with hardware virtualisation – allowing two or more operating systems and applications to time-share a single core without interaction. In the firm’s nomenclature, each operating system + application pair is in a ‘container’ – of which there can be seven, plus a container to hold the over-arching hypervisor software.
A single 8XT-A could, for example, run the infotainemnt system, the instrument cluster, a head-up display, and contribute an ADAS function, according to Johnson – some car manufacturers, he said, are talking about up to 13 screens in a car.
According to Johnson, the 100Gflops needed for driving assistance up to pedestrian prediction (see diagram right) is available from the 8XT-A architecture – and the flexibility offered by a GPU with general-purpose number-crunching extensions suits the current state of play, where algorithms have yet to settle, he said.
Beyond this, far more processing power will be needed for automated driving, multiplying by 10x for each self-driving level from 10Top/s at Level 3 to over 500Top/s at Level 5.
For this, Imagination is proposing to offer versions of its neural network processor.
“Neural networks are becoming a fundamental element in the development of automotive, from the occupant’s experience, through to ADAS and autonomous driving,” said the firm. “The PowerVR Series2NX neural network accelerator [NNA] could address applications such as driver monitoring, driver gaze tracking, seat occupancy, road-sign detection, drivable path analysis, road user detection and driver recognition. PowerVR’s NNA technology will play a role in enabling the industry to move from Level 3 to Level 5 autonomous driving.”
PowerVR Series8XT-A GPU is available for licensing now, then next-generation PowerVR GPUs, and NNAs for automotive, will be next year.
Although UK-based, Imagination Technologies was acquired in 2017 by California-headquartered private equity fund Canyon Bridge.