Get email delivery of the Cadence blog featured here
Remember automobiles from a few years ago? They were pretty simple machines with lots of small microprocessors distributed across the whole vehicle and doing fairly simple tasks. Semiconductors have been in automobiles for several years, of course, with simple 8- and 16-bit microcontrollers performing functions like Anti-lock Braking (ABS).
But today’s automobiles are a different beast. They incorporate several compute-heavy applications like Advanced Driver Assistance Systems (ADAS) systems, infotainment systems, and in some cases, even the ultimate goal that everyone is working toward—autonomous driving. All these applications need very heavy computing. For instance, in the case of ADAS, the system will need to process several images in real time, analyze them and make crucial decisions accordingly. The vision sub-system of the ADAS will need to process images from a variety of sources—RADAR, LIDAR, or standard image sensors. These pictures could be of speed signs, obstacles on the road, etc. Depending on the decisions arrived at after processing these images, the ADAS will need to control the steering and braking in a timely manner in order to achieve safe driving.
All of these processes take teraflops of computing. In order to feed this level of performance, we need high performance SoCs in leading-edge process nodes. And we need high-performance memory attached to them. This is a very different dynamic from what we have seen historically.
So what are the memory requirements for the automotive market?
Consider a SoC going into an ADAS. In many ways, it looks very similar to a mobile SoC. It has a processor complex in it, with a Flash interfaces, a DRAM interface or an LPDRAM, interfaces to some external image sensors, etc. So at first blush, it would seem that a lot of the memory requirements are actually quite similar between the two applications. Typically, it’s one or two devices on a module, right next to the SoC. It’s not a large memory subsystem like in a server farm. It’s also very cost sensitive. LPDDR seems to naturally suggest itself, thanks to the high-value DRAMS that a mobile market drives.
But the similarities end there. A slightly closer look will reveal a lot of a things very different between these two markets.
1. Operational Lifetime
Cell phones are notorious for needing to be replaced every 2-3 years for a variety of reasons, whether it be an unfortunate turn in a washer or an unfortunate drop onto a hard surface. As a result, a long operational lifetime of the device is not a very high priority in a mobile device. A car on the other hand typically lasts about 10 years. So the memory devices is a car definitely need a longer operational lifetime than those in a cell phone.
2. Long Lifetime Supply
Cell phone manufacturers usually replace phones every year or two with a new model and the market gravitates toward the latest model. This means that the requirement for availability of replacement parts for these cell phones is not very long. A car manufacturer, on the other hand, requires availability of replacement parts for almost 10 years, because even after they stop selling a particular model, if the electronics in the car breaks down, their customers still need to be able to fix them. So the automotive memories require a much longer lifetime supply.
3. Reliability and Safety
In a cell phone, if there is an issue, the worst that can happen is the phone needing a reboot. It is not a mission critical application. And definitely, no lives are at stake if the memory in a cell phone gets corrupted. So ECC, parity or bit errors is not a major consideration in a mobile LPDDR. In an ADAS in a car, on the other hand, thanks to the new safety standards, the electronics are being used for things that control the car and try to prevent accidents, if not to drive the car completely independently. The reliability and safety features within these devices are critical. So in an automobile, when it is a high reliability application, the memory absolutely must have ECC. It not only needs to be able to detect errors, but it must also tolerate some level of error.
So, while it’s true that the LPDDR4 memory interface standard is the same for both Mobile and Automotive applications, the actual devices that gets utilized because of the operating temperature required, and the lifetime of supply or reliability are very different, and the memory industry is going to respond to that change. Cadence is committed to making automotive designs safer, more connected and entertaining in less time and has a very comprehensive IP solutions for automotive memory applications and other automotive applications such as ADAS, Infotainment, Active Noise Control and Vehicle-to-Vehicle and Vehicle-to-Infrastructure. Please visit us to find out more.