Meeting Storage Needs for Software-defined Vehicles and Next Generation Zonal Architectures

Electronic systems in today’s vehicles are capable of otherworldly technological feats when compared to the cars we were driven to school in. In just a few years from now, the same will be true all over again. The rise of ADAS/AD vehicles along with more immersive In-Vehicle Infotainment (IVI) systems requires new thinking and new solutions under the hood. Right now, automotive OEMs are under tremendous pressure to reduce the complexity of electronic systems while increasing performance, two concepts that are not always in perfect alignment.

The shift from domain to centralized architectures, which was covered in a previous Tech Blog post, will require innovations in processing and memory, and a newly engineered storage device capable of supporting multiple hosts. All this needs to happen fast, too, because OEMs are expected to adopt zonal architecture at a 47% CAGR, reaching approximately 25M vehicles by 2030[i]. Zonal architecture promises to deliver the performance needed to usher in the next generation automotive systems.

A new automotive SSD design will need to meet the requirements of zonal architecture, but the story doesn’t end there. Capacity must expand in order to support increasingly demanding ADAS/AD and IVI systems. Data generation will soar as the number of sensors (RADAR, LiDAR, sonar, and cameras) per vehicle increases and the resolution of these sensors becomes higher. Operating systems and application code from gaming to streaming and video conferencing will also play a role in driving capacity growth. On top of that, security and serviceability are growing concerns. No one argues that centralizing storage doesn’t make sense, but the demands are higher than ever before.

Let’s look at some key concerns in designing a new, centralized automotive storage device:

Performance

In order to consolidate storage in zonal architecture, a new automotive SSD will need to provide high performance operation to minimize latency when servicing multiple hosts and images. PCIe is emerging as the leading SSD interface in next-generation automotive SOC’s for high-capacity and high-performance storage. While Gen 5 PCIe devices are just beginning to emerge in Data Centers and servers, PCIe Gen 3 is currently what’s in production today in the automotive industry. A next generation automotive SSD will benefit from PCIe Gen 4, which effectively doubles the throughput over PCIe Gen 3.

Throughput by Generation – PCIe Gen 3 & 4

Another factor that affects SSD performance is whether or not the drive contains internal DRAM. When writing to the SSD, the PCIe interface can transfer data from the host to the SSD faster than the data can be stored in the NAND memory. SSDs with DRAM buffer the write data so the ‘effective’ write throughput is higher. With this in mind, a new centralized automotive SSD should have internal DRAM in order to increase throughput and overall performance, improving upon current BGA SSDs, which do not contain DRAM.

I/O Virtualization

In a zonal architecture, SSDs will become a shared resource for multiple SOCs and will need to support more datacenter-like features. Various onboard systems will store content on the drive including downloaded content (gaming, streaming apps), maps for navigation, data collected from ADAS/AD systems and driver monitoring, and new applications as they become available, to name a few. This is a departure from how SSDs in domain architectures are used, where only a single SOC reads or writes to the SSD. A new zonal architecture SSD will need to simultaneously support multiple I/O hosts (SOCs) supporting multiple system images.

Virtualization Block Diagram

The above diagram shows multiple hosts connected to an SR-IOV enabled SSD via PCIe Switch. The SR-IOV enabled SSD will ensure that storage resources are sufficiently distributed while providing flexibility and sustained performance.

Form Factor

One of the biggest challenges in designing a new automotive SSD is finding a form factor that will meet the requirements of zonal architectures. Currently, automotive OEMs primarily use BGA-based SSDs due to the higher board-level reliability they provide. Soldered-down components such as BGA’s can withstand the higher shock and vibration tolerances that electronic boards experience in an automotive environment. BGA SSD solutions also provide the smallest footprint available. In a centralized architecture, where multiple SOC’s access a shared SSD resource, automotive OEMs will need to ensure the SSD capacity is sized for future growth over the life of the vehicle. 

It is estimated that a connected vehicle will generate 1.4TB to 19TB per hour[iii] with fully autonomous vehicles generating something in the realm of 20x more data. Some of this data will be stored to use in training to improve autonomous driving inferences. With Level 2 and Level 3 autonomous vehicles expected to reach almost 75% of market share in just a couple years from now, and as boot code, OS, applications and data become consolidated, it’s easy to envision capacities exceeding 4TB being needed. “For this reason, other non-BGA SSD form factors that offer capacities in this range today and offer the ability to scale to even higher capacities in the future should be considered,” says Vishal Devadiya, Senior Manager, Business Enablement at Samsung.

Passenger Vehicle Sales by Level of Automation

Source: BloombergNEF[ii]

Zonal architecture SSD requirements are closer to that of data center and server SSDs, where higher capacities are the norm, and a socketed system is used versus BGA. Two popular form factors are M.2, an older form factor which supports PCIe while also maintaining legacy SATA support, and E1.S, a newer form factor which supports PCIe. E1.S accomplishes a couple things that M.2 doesn’t: It features hot-swapability and its enclosure is designed with a heat sink to provide better thermal characteristics. M.2 and E1.S both support multiple lengths with longer lengths enabling higher capacities and both plug-in to a socket which allows for easy replacement. These types of SSDs are referred to as ‘rulers’ in the industry due to their flat, elongated shape.

The main concern with socketed systems in automotive applications is reliability due to shock and vibration. Automotive grade sockets already exist and can be developed to support a M.2 or E1.S based form factor that will achieve equivalent board-level reliability to BGA SSD’s. In the unlikely event that a socketed E1.S (or M.2) SSD does fail, it can simply be replaced - something that is very difficult to accomplish with a soldered down BGA SSD.

Power Efficiency

Capacity and performance versus power consumption is always a primary concern when developing a new SSD. This is true especially in automotive applications and even more so in EV’s, where OEMs want to meet a specific power budget to keep the battery size the same while also maximizing range.

If a socketed form factor is to be used to make gains in capacity and performance, we need to consider the power consumption implications. As shown in the table below, M.2 and E1.S SSDs consume more power compared to BGA SSDs due to the higher power supply voltages and usage of DRAM for buffering. While an increase in power consumption is not desirable, in a centralized architecture one SSD supporting I/O virtualization can replace multiple storage devices (found in domain systems today) thereby minimizing any increase in overall power consumption.

Security

There are now countless ways for thieves and hackers to access vehicles, either remotely or directly. On board computer systems for drivetrain, safety systems, IVI systems, bluetooth, wireless, and GPS, all expose vulnerabilities. Even something as familiar as a headlight can offer thieves access to systems by breaking through and establishing a direct wiring connection[iv]. The problem is so serious that cybersecurity for automotive is expected to grow at a 23% CAGR reaching $17.7B by 2031[v]. 

So, a new M.2 or E1.S inspired automotive SSD will need to be designed with the most nefarious intentions in mind. A socketed SSD raises a new concern in that it can be physically removed from the vehicle. “To address this, these SSD’s will need to ensure a method of attestation where only the data stored on the device can be accessed by the system in the vehicle it was originally installed in (or by the manufacturer)”, says Vishal Devadiya, Senior Manager, Business Enablement at Samsung.’

A New Automotive SSD Takes the Lead

Vehicles with zonal architectures will soon be on the road, with most OEMs expected to begin offering models in 2025/2026. Given the automotive development timeline of 18-24 months, there’s not much time to adopt a new automotive SSD that is ready to perform at the highest standards. So, can it be done? 

Over the past 2 years, the Samsung automotive memory team has been hard at work developing a new automotive SSD to help OEMs bring zonal architectures to market. Simply named “Detachable AutoSSD” while in development, it allows data access to multiple SoCs through virtualization. Earlier this year, Samsung presented a first ever of its kind prototype at Samsung Tech Day and again at Flash Memory Summit (FMS). Once in production, Detachable AutoSSD is planned to ship in an E1.S-like configuration with support for PCIe Gen 4 connectivity. 

As with the rest of Samsung’s automotive memory lineup, Detachable AutoSSD will meet the highest standards for quality, performance, and reliability. The entire device, including the NAND, DRAM, and controller, will be manufactured in Samsung’s FABs. Samsung’s commitment to providing automotive OEMs with outstanding products continues with a cutting edge Detachable AutoSSD for zonal architectures. Stay tuned as the story unfolds…