For mission-critical applications where failure is not an option, fail-safe memory with embedded functions will revolutionize the design of highly reliable and secure systems.
The first microprocessor to integrate memory revolutionized the industry. Direct internal access to memory increased throughput and reliability while eliminating points of failure and reducing system cost. Today, integrating several types of memory into a microprocessor adds real value, depending on the application. This trend is now reversing: memory is beginning to integrate microprocessors, which will again revolutionize the electronics industry and bring a level of smarts to applications requiring reliability, monitoring, security and artificial intelligence.
Mission-critical applications, where failure is not an option, use memory storage devices to safely boot up the system, log critical information, and extend working memory for critical functions. Examples of such “fail-safe” applications are advanced driver assistance systems (ADAS), portable medical devices, factory automation, defense-level sensors, and advanced wireless communication systems. Each of these applications contain mission-critical capabilities or subsystems that cannot be allowed to fail. In these systems, failure often comes at a high price, such as damage to the system, expensive down time, and/or injury or death.
From a memory standpoint, fail-safe operation means 1) the system can have confidence that when a value is stored in memory, the same value will later be retrieved, and 2) that the system will know if a particular value is no longer reliable so it can take corrective action. While shrinking process technologies have improved memory density, they have also increased the risk of memory corruption, such as the bit-flips caused by cosmic particles. Thus, memory reliability is quickly becoming a standard concern of OEMs across all mission-critical applications, but especially in industries where safety compliance regulations and always-on criteria must be met.
All fail-safe systems need to accurately store boot code and to start up reliably from that code. As cars become more autonomous, the “thinking” part of the car needs accurate sensor data to make its decisions. Data errors could result in the misidentification of hazards. Thus, more of the system memory needs to provide high reliability to guarantee fail-safe operation.
To increase memory reliability, the host processor will compute an error correcting code (ECC) value and store this value so the system can later verify that data read back is the same as when it was stored. ECC also allows the system to potentially correct errors. However, the ECC value takes computing resources to calculate. The system must also verify that the ECC values weren’t bit-flipped when they were stored.
Another example of a high reliability task is wear-leveling for nonvolatile memory (NVM). Because NVM cells have limited endurance, systems with frequent writes must spread cell use and wear across the memory to prevent premature loss of memory blocks. Implementing effective wear-leveling puts an additional burden on the host processor.
Today’s most-advanced memory products have been optimized to meet the needs of mission-critical applications, meaning they are specifically designed to store, protect, and restore data under intense workloads and in harsh environments. These enhanced memory devices feature integrated compute capabilities to achieve high levels of functional safety, security, and reliability. Integrated computing enables additional functionality—for example, to serve as a root of trust.
Rather than relying on the host processor to assure memory reliability, fail-safe memory devices manage this function themselves. This enables important capabilities such as instant-on functionality, since boot code can be verified and loaded faster than if the host processor is required to perform such verification. Additionally, the integrity of configuration and other important data can be verified quickly and reliably in fail-safe memory devices. Advanced memory devices can also monitor their own operation to assure integrity of reads and writes between the host and memory device.
Time-to-market is also substantially improved when using fail-safe memory with embedded functions. Designing high reliability capabilities for an application with safety requirements takes time. It also requires extensive analysis to verify compliance with regulations. Choosing a fail-safe memory with these integrated capabilities can save both time and money.
Eliminating the need to specify a higher performance host processor for verifying memory reliability also generates additional cost savings. By integrating necessary processing capabilities into memory, reduced total cost-of-ownership is possible with the next generation of smart memories.
Fail-safe memory spans all types of memory, from SRAM to nonvolatile devices including NOR Flash, F-RAM, and NVRAM. For example, NVRAM can be used for data logging or making trajectory calculations for an autonomous vehicle also needs to be just as reliable as the NVM that stores the application code used to make the calculations.
The Future of Memory
While the primary benefit of smart memory is higher reliability, there is a secondary benefit that will become increasingly important: a faster pace of innovation. Consider how fast safety regulations are changing and the need for security to continuously address new threats. With an integrated memory plus processor architecture, OEMs have the ability to implement new safety and security measures faster than if these were implemented in software. Additionally, when regulations can be met by a smart memory subsystem, existing applications can immediately accommodate these regulations. OEMs can also introduce new reliability features to product lines without requiring a redesign, simply by changing out the memory.
Fail-safe memory will also be able to take on the burden of securing data. By integrating cryptographic engines alongside the embedded memory processor, data can be stored in a secure manner. Given that the memory footprint typically dwarfs the number of gates required for a processor, these and other advanced capabilities can be implemented in memories at a relatively small cost.
Yet securing data is just the beginning. Imagine a memory that could perform calculations on sensor data before storing it. Fail-safe memories are also inherently more reliable because they are not dependent on an external processor.
The numbers of ways smart memory will offload host processing in the future is endless.
In the future, commodity memories will continue to play a role in the market, but for those applications where failure is not an option, fail-safe memory with embedded functions beyond just storing data will revolutionize how highly reliable and secure systems are designed.
>> This article was originally published on our sister site, EE Times.
Amr Elashmawi is vice president of marketing at Cypress Semiconductor.