An integral part of the Internet of Things (IoT), machine-to–machine (M2M) refers to the connectivity technologies that enable embedded computing applications, processors, sensors and other devices within the same communication infrastructure to communicate with each other with minimal human intervention. An M2M system solution includes the hardware, hosted software, and services necessary to easily develop applications. It also includes wired and wireless communication devices, modules, and adapters. In addition, an M2M solution provides management, messaging, and storage services that connect remote assets, enterprise applications, and much more.
In complex environments, M2M devices are stand-alone devices equipped with a specific interface (for example, USB), volatile and nonvolatile memory stack, applications, operating system, and device drivers. (See Figure 1.) Typically, M2M devices use some form of Flash memory for code storage and a volatile memory, such as SRAM or DRAM, for caching or buffer purposes. Because M2M services are so varied, their deployment requires the use of applications on the hosting device that are able to handle information and M2M communication and, in some cases, provide a secure communication channel for applications that deal with financial transactions like point-of-sale (POS) terminals.
These types of system environments add complexity to the system memory requirements. Memory technology matters, whether it’s used for storage or to enable the processing of new kinds of data to improve efficiencies and increase intelligence in systems. Several factors go into the decision-making process to select the right components, and one cannot be sacrificed for the other. A less expensive solution may not address longevity issues, which are a must for M2M products that have longer lifetimes. This article focuses on the challenges of selecting the right non-volatile memory (NVM) for M2M applications and some of the key factors to consider.
Flash Memory Considerations for Short-Range M2M Solutions
Short-range M2M solutions encompass sensors and devices within the same network infrastructure, such as a local area network (LAN) or a personal area network (PAN). Short-range solutions include an abundance of wireless communication technologies ranging from various proprietary options to standardized options such as Wi-Fi, Bluetooth, Zigbee, and Z-Wave. These standards will continue to evolve to keep up with the requirements of M2M communications and to support and ensure interoperability with the continuously growing list of applications.
Embedded memory dominates the landscape of short-range M2M solutions that employ Wi-Fi, Bluetooth, Zigbee, Z-Wave, and others. Embedded memory saves space, provides faster boot-up time, and helps improve Execute-in-Place (XiP) efficiency. However, there are disadvantages one has to account for in system development.
The biggest disadvantage is that embedded memory is lower density and poses the risk of running out of bits to accommodate the latest software revisions. It is, therefore, prudent to design-in external memory, such as a 128Mb Serial NOR or 1Gb–2Gb NAND Flash memory, to ensure there are enough bits for the program code and to maintain an image of the previous version in case of power failures.
Embedded memory is also limited in write cycles and may not be suitable for parametric storage, especially in applications that need to update system parameter data several times over the lifetime of the application. While this limitation can be overcome with cache memory, it comes at the expense of additional system costs. The final consideration is the need to accommodate for the increase in active system power for updates to the Flash and the hit to overall system efficiency during erase and write cycles.
Flash Memory Considerations for Long-Range M2M Solutions
Long-range M2M solutions require always-on connectivity for sensors and devices in large infrastructures that cross several network boundaries. Cellular technology is the most common application for these solutions. M2M cellular modules are available in 2G and 3G/4G versions. The 3G networks are based on the global High-Speed Downlink Packet Access (HSDPA) standard, which is compatible with common microprocessor unit (MPU) architectures and enables rich, real-time multimedia and browsing experiences. The 3G networks will continue to be a popular international choice over slower 2G networks and will continue to add subscribers.
The number of cellular-based IoT connections is expected to nearly double in the next three years. Cellular providers are focused on upgrading to 4G LTE networks to accommodate the high-capacity data services more efficiently. The mobile network operators (MNO) who are deploying revenue-generating M2M services are leveraging the 3G and 4G networks to utilize network resources. Consequently, M2M module suppliers are excluding 2G-certified modules from new designs to adapt to newer component solutions and state-of–the-art manufacturing. The 3G M2M modules are, therefore, in the sweet spot of market growth and are expected to occupy nearly half of the cellular M2M connections by 2016.
Performance, Density, Cost, and Packaging Considerations
Figure 3 shows a typical cellular M2M module, including a software stack. The Flash memory requirements range from 32Mb in 2G cellular modules to 4Gb in 4G cellular modules. The 2G cellular modules have slower modem speeds and have a smaller 32Mb/64Mb memory footprint for critical communication code. Here, NOR Flash is the memory of choice; its execute-in-place (XiP) architecture enables code execution out of the Flash memory and only requires 16Mb–32Mb of PSRAM in working memory or extended cache. The 3G/4G cellular modules have higher performance and density requirements. The critical communication code is much higher, and the system has to store a copy of the software image and Java middleware versions. This drives the density requirement to as high as 4Gb for NAND Flash memory. From a cost standpoint, SLC NAND is a good choice for 1Gb densities and higher.
Switching to NAND Flash will, however, mean that the code cannot be directly executed out of the Flash; instead, it will have to be downloaded into the RAM for execution. This is referred to as store and download (SnD), or compute memory architecture, and it causes the external DRAM requirements to increase significantly to shadow and execute code. Even with the increase in the DRAM, however, the overall memory solution will be in price parity with a NOR Flash and lower-density DRAM solution.
Owing to the small form factor of M2M modules, multichip packages (MCPs) are preferred because they offer significant savings in board space compared to separate Flash and DRAM components. Figure 5 illustrates an example of a 1Gb NAND/512Mb LPDRAM MCP solution in a 8 x 9 x 1mm package that fits well in a 3G/4G M2M module.
Product Longevity
NOR Flash memory has a better track record than NAND Flash in supporting automotive, industrial, and medical applications that are characterized by long lifecycles, and the average lifespan of a M2M module is typically 10+ year. NOR Flash is also self sufficient in handling all of the memory functions, whereas NAND is dependent on the processor for managing the memory functions, including ECC, bad block management, and wear leveling. This can make it difficult to extend support over the product lifetime and through NAND technology shrinks because it is difficult to ensure that the NAND characteristics will remain unchanged through geometry changes.
Use Case Scenarios
The capability of in-system memory updates is one of the key advantages that Flash memory offers to embedded applications that typically require approximately 100,000 PROGRAM and ERASE cycles and a minimum of 10 years of data retention. NOR Flash memory has traditionally met this requirement through several generations of shrinking geometries. In the case of NAND Flash memory, characteristics change with shrinking lithography. The effect is even more pronounced in technologies below 30nm geometry, as shown in Figure 6, where ECC requirements increase significantly and PROGRAM and ERASE cycles become limited for embedded applications.
The use case scenario becomes a critical factor in choosing the Flash memory that is best suited for a specific application. For example, applications that employ SnD architecture, where the code is copied into DRAM during power-up and updates are infrequent, can exploit the cost advantages of NAND Flash, especially when the code density requirement exceeds 1Gb of memory. On the other hand, a more careful assessment of the use case scenario is needed for M2M applications where frequent code and data updates will drive up the PROGRAM and ERASE cycle requirements beyond the technology limitations.
Conclusion
M2M devices are connecting new places—such as manufacturing floors, energy grids, healthcare facilities, and transportation systems—to the Internet. When an object can represent itself digitally, it can be controlled from anywhere. This connectivity means more data, gathered from more places, with more ways to increase efficiency and improve safety and security. The increased data demands exponentially drive the bits required for storage and executing actions as required by the data. Performance, density, cost, packaging, longevity, use case scenarios are key factors in choosing the right memory solution for your M2M products.