What it is all about: DRAM memory as a challenge for modern automotive systems
The amount of data that needs to be processed in real time in modern vehicles is constantly increasing. At the same time, there are requirements for low power consumption as well as great cost pressure, which leads to the increased use of components originally developed for the consumer market. The result are heterogeneous computing platforms consisting of GPUs, dedicated accelerators, CPUs, as well as DRAM memories, in particular.
Current ADAS applications require a DRAM bandwidth of approx. 100 GB/s. This requirement can currently still be met easily with memories developed for smartphones and tablets (LPDDR). In the next few years, the transition to autonomy levels 4 and 5 will increase the required bandwidth to up to 400-1000 GB/s, which can only be implemented in the long term with graphics memories (GDDR or HBM). In order to meet the safety and security requirements, diagnostic mechanisms such as memory tests or error correction mechanisms (ECC) must be used. Thus, the use of DRAMs for autonomous driving also poses major challenges in terms of safety and security. It is therefore important to address these challenges at an early stage and to pay close attention to all constraints.