The amount of software and data used by more sophisticated digital systems is likely to surpass the capabilities of flash embedded on microcontrollers, forcing engineers to add discrete flash memory [that which stores data in the absence of a power supply and can be electrically erased and programmed] to the system’s bill of materials. These flash chips may be needed for many applications: storing the code that powers autonomous systems, holding safety sensor data that precedes crashes and supporting infotainment clusters.
Microcontrollers now routinely hold 4 to 8 Gbytes of embedded flash, enough to handle the needs of many systems. Some radio head units and clusters already add some discrete flash, but capacities are low. More storage will be needed for safety systems that acquire raw input from cameras and radar, and autonomous vehicles will need to store voluminous amounts of control software. Flash usage is expected to soar.
“The typical vehicle today probably has around 10 Gbytes of storage space, though some high-end vehicles use hard disk drives so their storage capacity can go up to 64 Gbytes,” said Martin Booth, Product Marketing Director at Western Digital. “By 2020, it’s easy to see the upper end going up around a terabyte and an average vehicle going into the hundreds of Gbytes. Most storage today is in infotainment, but by 2020 other areas will use as much storage as infotainment.”
As advanced driver assistance systems (ADAS) perform more active braking and steering, insurance companies and manufacturers may want to store data that details what was happening seconds before a collision. That may force engineers to deploy flash on the circuit board or in the package that holds the CPU.
“In ADAS, some suppliers may do a system in package with on-board flash,” said Nina Turner, Automotive and Energy Research Manager at IDC Research. “Systems may store data around an event, an accident or something. The question is, how much traceability is needed? Is 30 seconds too much or not enough?"
Turner expects additional memory requirements in ADAS, especially when companies use high-volume raw data from sensors. "There’s got to be someplace to store the input while data is being processed,” she noted.
Buffering may become a central need for autonomous vehicles, opening the door for DRAM as well as for flash. Along with raw data from cameras, radar and lidar, communications from other vehicles and roadside infrastructure stations will often need to be queued up for processors. Flash may also be needed to buffer over the air updates. The soaring volume of software in cars is another factor driving growth.
“By 2020, there may be 300 million lines of code stored on a vehicle,” said Robert Bielby, Senior Director of Automotive System Architecture at Micron Technology. “When you look at autonomous cars, the sheer quantity of software needed for a vehicle to drive down the road is very high, so there’s a huge demand for memory.”
While some data will be stored in the cloud, many technologists say that’s not viable for many types of information. It takes a lot of bandwidth to constantly stream data to the cloud using connections that aren’t free. Additionally, anything that’s safety related needs very low latency and very high reliability, traits that cellular links can’t necessarily guarantee.
IDC's Turner predicts that memory will grow at about the same rate as automotive semiconductors, around 9% compounded annually. She also predicts that most chips will be embedded.
Using some form of removable memory card adds security risks, since hackers could write malware and easily get it into vehicle electronics, she noted.