Storing data in the IoT
05 October 2016
Data farms for the Cloud are hugely memory intensive, but there is also demand for the Things to store increasing amount of data locally.
(Click here to view article in digital issue)
The requirements of IoT data and applications highlight some flaws and compromises inherent in many of the ‘traditional’ memory technologies.
Whilst the selection criteria for Cloud data storage focus on physical size, reliability, speed and power consumption, data storage at IoT nodes has different priorities. Embedded flash (eFlash) memory is well suited to IoT applications requiring storage of critical data and code. Its field programming capability gives the flexibility to store last minute system-level changes, whilst relatively high performance (>6.4 GB/s readout speed) and high density enable typical embedded microcontroller applications.
As the number of IoT devices in the field grows exponentially, so does the requirement for greater intelligence in the node along with larger storage density and the needs for integrity and security of data stored there. NOR flash memory is typically used in such embedded applications. However, in order to deliver the extra functionality in IoT devices, demand is growing for larger memory densities for saving software (including boot up programs, firmware, and embedded OS) and data (including log data). This is driving demand for single-level cell (SLC) NAND flash memory, which offers higher density and comparable reliability than NOR flash memory - and at lower cost per-bit.
These characteristics are embodied in new devices like Toshiba’s Serial Interface NAND IC, which offers three densities up to 4Gb based on 24nm technology. Compatibility with the widely used Serial Peripheral Interface (SPI), which can be controlled with just six pins, allows the new “Serial Interface NAND” to be used as SLC NAND flash memory, with a low pin count, small package and large capacity.
Improving integrity & security
A disadvantage of standard SLC NAND Flash memory is that error correction code has to be provided by the host processor. Building ECC into the memory chip hardware itself removes that burden. BENAND devices with hardware ECC are available in densities from 1Gbit to 8Gbit, providing all the necessary endurance and data retention for sensitive or frequently used data in IoT nodes.
Greater security and even better performance are exemplified by STMicro’s IC designed to complement a range of IoT-capable microcontrollers. Devices such as the STM32L476xx family embed high-speed memories (Flash memory up to 1 Mbyte with up to 128 Kbyte of SRAM), a flexible external memory controller (FSMC) for static memories, a Quad SPI flash memories interface and an extensive range of enhanced I/O and peripherals connected to two APB buses, two AHB buses and a 32-bit multi-AHB bus matrix. These memories embed several protection mechanisms for embedded Flash memory and SRAM including readout protection, write protection, proprietary code readout protection and Firewalls.
Higher storage capacity
Meeting the need for higher data storage densities in IoT, embedded Multi Media Card (eMMC) standards provide the equivalent of familiar MMC cards on a chip. Features of the e-MMC are low cost, high performance, and low power consumption. Since it has a specifically designed controller, it can be easily integrated into application systems which have a MMC/HS-MMC interface. The latest JEDEC eMMC Version 5.1 defines two new features to improve performance and security: command queuing and secure write protection.
Command queuing allows users to process multiple tasks generated by the user's issue of multiple commands, in the order of the user's preference, by initially storing the tasks in a waiting queue. Toshiba has quantified its benefits, finding that command queuing improves random read performance speed by approximately 30% at maximum compared to existing products.
Secure write protection expands the conventional write protect feature and protects the user's data stored in an assigned area from being overwritten or erased by unauthenticated users.
Products based on these latest eMMC standards integrate NAND chips fabricated with a controller to manage basic control functions for NAND applications in a single package. A fast HS-MMC interface is included and typical capacities available run from 16GB, 32GB and 64GB to 128GB.
Running out of space? Go 3D
Where higher performance is required eMMC-based solid-state storage just isn’t as fast or robust as a full solid-state drive. New 3D Flash memory technology could change all that. In contrast to earlier NAND Flash memory, where cells are formed on a two-dimensional silicon substrate, this new technology involves stacking Flash memory cells vertically on a silicon substrate to give significant density improvements.
A number of silicon vendors are beginning to manufacture 3D NAND memories, and several solid state disk drives based on these ICs have been announced already. For example, Toshiba announced the first prototype 3D flash memory technology in June 2007; the company is actively promoting BiCS FLASH to meet demand for larger capacity with smaller size. The company’s latest device incorporates three-bits-per-cell (triple-level cell, TLC) technology and achieves a 256-gigabit (32 gigabytes) capacity.
As BiCS FLASH technology gets more refined, the next milestone on the development roadmap is a 512-gigabit (64-gigabytes) device, also with 64 layers. The 64-layer stacking process enables 40% larger capacity per unit chip size than 48-layer stacking process, reduces the cost per bit, and increases the manufacturability of memory capacity per silicon wafer. Upcoming 64-layer BiCS Flash can meet demanding performance specs; so the early beneficiaries will probably be data centres, where SSD modules can provide massive capacity in a small space with low power requirements.
Storage based on 3D NAND Flash will be useful for applications that include enterprise and consumer SSD, smartphones, tablets and memory cards as well as IoT nodes.
MRAM promises zero-power standby, instant-on
Another area where the IoT may benefit from newer memory techniques is the emerging technology of MRAM - Magnetoresistive random-access memory. Instead of storing data as electric charge or current flows, data in MRAM is stored by magnetic storage elements formed from two ferromagnetic plates separated by a thin insulating layer. One of the two plates is a permanent magnet set to a particular polarity; the other plate's magnetization can be changed to match that of an external field to store memory. This so-called magnetic tunnel junction is the basic building block of an MRAM bit. MRAM devices are built from a grid of these blocks.
There are a number of techniques for writing data to MRAM cells. The simplest "classic" design, places each cell between a pair of write lines arranged at right angles to each other, parallel to the cell, one above and one below the cell. When current is passed through them, an induced magnetic field is created at the junction, which the writable plate picks up. This approach requires a fairly substantial current to generate the field, however, which makes it less-than-ideal for low-power applications.
A newer technique, spin transfer torque (STT) or spin transfer switching, uses spin-aligned ("polarised") electrons to directly torque the domains. Specifically, if the electrons flowing into a layer have to change their spin, this will develop a torque that will be transferred to the nearby layer. This reduces the amount of current needed to write the cells, making it about the same as the read process.
As a fast-write, non-volatile memory, MRAM potentially has many advantages in support of the Internet of Things. For one: many IoT applications operate in intermittent access or batch mode, making MRAM the perfect solution where fast-write working memory data must be continually updated yet preserved between batch read accesses or when power is lost.
Another advantage is where there are long or frequent periods of standby, yet instant-on is required. MRAM enables instant access to data as well as critical code. In addition, it preserves data for up to 20 years, yet retains very fast write times.
Finally, in IoT nodes where devices spend a lot of time in deep sleep mode, MRAM can be powered down completely with zero energy consumption, yet data are non-volatile and MRAM has a very fast power-up write time.
MRAM pioneers Everspin reckon that over 50million of its MRAM and ST-MRAM products are currently deployed in datacentre and cloud storage facilities and across energy, industrial, automotive, and transportation markets. OEMs building IoT nodes have a choice of parallel, SPI or DDR3 interfaces, at densities from 128kb to 64Mb.
Technologies like 3D NAND Flash and MRAM may be in their infancy, but could deliver substantial benefits for IoT data storage. Meanwhile, significant improvements in more conventional Flash memories are meeting current demands for data storage in globally connected Things.
Contact Details and Archive...