7 years later, what we got right and wrong about autonomous cars

“Five stages of automation infographic”

Verdict: Right but updates are needed

Besides the brand color scheme and old Micron logo, this infographic on the five stages of automation is still accurate and relevant today, (and is a nostalgic throwback for me). We didn’t coin the five stages of autonomy; it's an industry standard established by the National Highway Traffic Safety Administration (NHTSA). The major update needed for this visual is the "today" position. Many new vehicles are now in the " Partially automated" stage, with some advanced electric vehicles (EV) capable of stage four autonomy. Stage five technology exists and is seen in limited cities through pilot programs, though they are generally limited to warm and predictable climates. To be considered truly at Stage five, a vehicle would have to be able to handle any situation and climate condition that it could encounter. So, why aren’t all new cars fully autonomous? The list of obstacles is extensive and includes government regulation, safety and total cost of ownership. Perhaps we will be at Stage five within the next seven years. That possibility draws closer as Micron continues to make groundbreaking innovations in memory bandwidth and in centralized automotive storage architectures. Memory bandwidth and centralized architectures enhance performance, reliability, and safety, all of which are critical for achieving Stage five autonomy.

“Memory, the unsung hero in autonomous driving” and “high-speed memory is an essential component of autonomous driving”

Verdict: Right and as true as ever

Memory and storage technologies are crucial for autonomous vehicles at any level of autonomy. AI in cars must handle vast amounts of data to make quick decisions at high speeds. For autonomous driving, excessive delays and latency are unacceptable. Humans are pretty good at making snap decisions, despite what we may observe in traffic. The problem is, we get distracted easily. Computers don’t. Cameras, lidar and sensors are always vigilant. High-performance vehicle computing platforms don't nod off behind the wheel and are always calculating and making decisions to keep everyone safe, as long as they get the data they need from memory and automotive storage. High bandwidth memory and centralized storage solutions are crucial for tackling these data challenges and keeping the system running smoothly, efficiently and safely.

Memory is also essential for complex AI training models and for enabling rapid inference responses. Micron’s innovations provide the reliable foundation needed for these demanding AI tasks. By making reliable and efficient memory and storage, Micron plays a key role in integrating AI into vehicles, advancing autonomous driving capabilities.

“The importance of GDDR6 in the future of autonomous driving”

Verdict: Wrong

At the time, Micron’s GDDR5X memory was being used in automotive solutions. GDDR5X and GDDR6, primarily for gaming, also entered automotive solutions and networking applications due to their high-speed capabilities. Today’s graphics and AI applications are using GDDR7, the latest graphics memory standard. The expectation was that this trend would continue in automotive solutions, especially in infotainment systems, where screens are getting larger and offering higher resolutions. Vehicles like the Tesla Model 3 and Cybertruck even offer an in-car gaming feature called Telsa Arcade, where video games can be played on the infotainment system using the pedals and steering wheel (not while driving for safety reasons). So why did we get this wrong? It seems logical that the need for more bandwidth and graphics memory would grow, right?

The need for more memory and bandwidth continues, and increases with each level of autonomy, but today the type has changed to low-power double data rate (LPDDR) memory. Originally built for low power consumption in mobile phones, LPDDR, specifically LPDDR5X, became the best fit for modern vehicles.

Mobile phones and cars share many similarities: they don’t stay connected to a power outlet, they run on batteries and they are full of sensors and compute that constantly gather and monitor data. AI interactions have become expected from users. Micron continues to innovate low-power memory solutions to increase performance and lower power consumption to meet these expectations.

One recent example is the introduction of LPDDR5X DRAM with direct link ECC protocol (DLEP). This LPDDR5X-optimized error correction code (ECC) scheme mitigates all system in-line ECC which helps to deliver a 15% to 25%¹ bandwidth increase. DLEP not only delivers increased performance but also helps LPDDR5X memory systems achieve the ISO 26262 ASIL-D hardware metric through reduced failures in time (FIT). Additionally, this new product delivers approximately 10% lower power consumption on a pJ/b (picojoule-per-bit) perspective and a minimum 6% additional addressable memory space². Built on the backbone of Micron’s certified ISO 26262 ASIL-D systematic LPDDR5X DRAM, critical automotive functional safety (FuSa) requirements are now readily achievable. Simply put, DLEP increases bandwidth and reduces power consumption, the holy grail of win-win solutions for autonomous driving workloads.

There have also been many advancements in data storage that were not foreseen. Centralized storage SSDs like the Micron 4150AT SSD are completely changing automotive storage architectures. Rather than having one storage chip for infotainment data, another for ADAS systems and more for seat positions, this architecture consolidates data storage from up to four systems into one robust SSD that is certified to withstand the rigorous needs of vehicles. This effort, combined with AI, allows for an increased number of automotive storage architectures for vehicle applications.

But what type of memory comes next for vehicles? Will LPDDR continue for the next seven years, or will AI high-bandwidth memory (HBM) like Micron’s HBM3E move into vehicles? That is likely to happen, especially as AI becomes more prevalent and necessary for fully autonomous driving, but perhaps it will be HBM4 or HBM4E that first makes its way into autonomous vehicles. I’ll give one parting prediction: It won’t take seven years for a Micron HBM product to be adopted into an automotive solution.

Conclusion

I won’t start keeping score of what predictions I’ve got right and wrong (I’m not sure my ego could handle that), but it is a useful exercise to gauge our perspective on future technologies and trends. The number one takeaway is that in the past seven years, the focus and value of AI in vehicles has increased exponentially and the importance of high-performance memory and storage solutions for AI-powered vehicles has done the same. The specific memory and storage solutions may change, evolve and improve, but the challenge they solve does not. Data is at the core of all AI, and data lives in Micron memory and storage solutions. But our technology doesn't just store data; it accelerates the transformation of data into actionable intelligence.