For the father of Qualcomm’s automotive division Nakul Duggal, the number of chips in cars and their power will increase tenfold in the next decade. A rise in power necessary for the growing complexity of the calculation tasks that the vehicles of the future will have to solve.
The quest for autonomous driving brings with it a huge challenge: that of transforming our vehicles into veritable little supercomputers. This transformation from the “dumb” car to the “intelligent” car is whetting the appetite of world champions in semiconductors, ready to sell their chips to a market whose demand for computing power is exploding. Among the companies that intend to carve out the lion’s share, there is the American Qualcomm. Known for its world number 1 position in mobile processors and other modems, the American has extended its ambitions to the connected and autonomous car. Thanks to its know-how in telecoms on the one hand, and in the design of smartphone chips, on the other.
” Within ten years, the most high-end models of cars will have a combined computing power of their chips ranging from x7 to x10 compared to today’s best vehicles. “, predicts Nakul Duggal, vice president and head of the automotive division of Qualcomm. A gain in power necessary with the evolution of uses. ” In the past, there was a backup camera in your car. And its only use was to be called by an application that projects the video stream on the central screen. But now, several cameras, sometimes different, are controlled. In addition to simple display, cameras are also used for piloting, detection and security purposes. continues Mr. Duggal. ” The need for processing power is of a whole different magnitude! »
The number of sensors, cameras and others, being in full explosion in our vehicles – electronics should represent 20% of the cost of cars by 2030 – and the uses being more and more complex, it is therefore necessary to design processors always more powerful and perfected. But unlike the world of smartphones where every passing year sees new chips arriving at perfectly regular intervals, the design of the chips in our cars is a whole different matter.
Chips extremely difficult to design
If Moore’s “Law” predicts a doubling in performance of computer chips every eighteen months, that of automotive chips in the next decade cannot be so fast. It is slowing down on one side because of the increasing difficulty in reducing the size of the circuits. An evil that will strike all of the cutting edge chips. But it’s also the very nature of automotive platforms, where safety (here understanding platform stability) matters more than anything, that limits raw power upscaling. It also implies adapted developments… and therefore slower.
Read also : How Semiconductors Will Soon Make Up 20% of Your Car’s Price… Hopefully (September 2021)
” Designing a chip for a car is extremely harder than for a phone “says Nakul Duggal. This industry veteran – he joined Qualcomm in 1995! – explain that “ the simultaneous calculations are not only more numerous, but in addition the usage scenarios are also more varied, because there are so many parameters to take into account and calculate. “. Added to this is the fact that auto processors have a much longer life. rock & roll than our phone chips. ” The automobile’s natural need for power causes chips to operate at a higher frequency than in your smartphone or PC for example “, explains Nakul Duggal. ” Between the calculation and the outside heat, they can sometimes operate at more than 105°C “, he continues. ” The processors must operate at all times, without the slightest interruption or breakdown. “. The chips must therefore be hyper resistant and with zero defects. And this is where industrial know-how helps.
The contribution of smartphones is not what we think
At first glance, one might think that automotive chips mainly benefit from a variation of different smartphone technologies – CPU, GPU, NPU, etc. This assertion is partly erroneous. ” We do not recover the same technological blocks as smartphones refutes Mr. Duggal. ” We actually don’t embed a phone GPU in the other chips for example. The different IP blocks (CPUs, GPUs, NPUs, ISPs, etc.) are now developed taking into account from the beginning of their design the different variations that we will need. For example, our NPU Hexagon is, from the start, designed to be suitable for different uses. In mobile, its power will be used for text prediction or photo enhancement. But, in the automobile, it is used for example for predicting the movements of vehicles. This must be taken into account from the start so that the functions are best suited to subsequent uses. »
This common development of the different “pieces” of the chip is therefore transverse between all of Qualcomm’s chip development units (smartphones, routers, PCs, cars, etc.) and turns out to be more horizontal than what we can read. here and there. However, smartphone chips do bring a critical asset to their automotive sibling: “ Qualcomm delivers about 100 million chips per quarter, the majority of them for smartphones, so in the nodes of production (the finesse of engraving, ndr) the most recent. Thanks to these huge volumes, we were able to establish production models to obtain good yields, detect problems and errors, and qualify the nodes the most recent “. An industrial strength that allows Qualcomm to use state-of-the-art engravings for its new Snapdragon Flex automotive platform.
State-of-the-art engraving finesse and concrete design
In addition to its production cost, any electronic chip must find a balance between three physical factors called PPA: energy power (Power), performance (P) and area used (Area). ” With a chip engraved in 28 nm, it is impossible for us to reach the same level of power as with current chips in 5 nm says Duggal. ” Ia chip would be too big and would heat up too much “. So there is no choice but to reduce the size of the chips. But isn’t the fineness of the circuits a source of problem as we often read?
” No way. As I was telling you, on the one hand, our very high chip production volume allows us to perfectly predict the final quality. But in addition, everything in automotive processors is designed for resistance and safety. How the logical elements are arranged (IP in the jargon), from the nature of the logic cells to the manufacturing processes, everything is validated and qualified to specifically meet automotive constraints “. This resistance, both physical and logical, is what allows Qualcomm to create an all-in-one chip with Snapdragon Flex. Asked about the risk of losing the redundancy of the multiple chips that today manage driving assistance, ABS, the navigation system, etc. he replies that ” this is also taken into account. Both from a hardware and a software perspective “.
Read also: Mobileye introduces EyeQ Ultra, the chip that promises true Level 4 autonomous driving in 2025 (January 2022)
This confidence in the performance and robustness of his solution, Mr. Duggal carries it to the hoped-for success of the platform. Asked who is Qualcomm’s biggest competitor in complete automotive platforms – Nvidia? Intel? – Mr. Duggal asserts that “nWe don’t really have any. We obviously have good competitors in certain areas, such as the very good Mobileye solution that you mentioned. But our networking know-how coupled with our ability to quickly design all-in-one chips is hard DNA to replicate. The future will tell if Qualcomm will seduce an automotive industry that is now condemned to digitize all the elements of its vehicles. Which will soon be more running computers than “smart cars”.