If the variety of new connected devices uploading gigabytes of data to the cloud shown at CES 2017 is any indication, the Internet of Things (IoT) will continue to create a greater demand for improved memory technology performance and power efficiency.
While technologies have come a long way to support a wide range of connected devices, from voice and vision-activated robots to gadgets such as fitness tracking rings, the energy budget of many IoT devices still remains a design challenge. An IoT device that needs to achieve a 10-year lifetime on a single CR2032 coin cell would require it to operate at an average system current of just 2.5 μA. By comparison, a device operating at 300 μA average current would require a new coin cell battery every month.
Devices with the ability to run on batteries for years are particularly critical in remote applications, such as the thousands or tens of thousands of smart sensors or other connected objects deployed across a wide area like a city where regular battery changes would be impractical if not impossible. These IoT devices may not even have a finite energy budget from a battery, and rely instead on energy harvesting derived from ambient sources such as solar, kinetic, or thermal sources.
Meeting IoT energy requirements demands a variety of energy saving strategies. Most IoT devices are designed to spend most of their time in standby or deep sleep mode, and activate only to perform necessary tasks before returning quickly to an inactive state; Indeed, the best way to save energy is to do nothing at all, but this also means there’s a requirement to get the most done during the time an IoT device is active.
For example, one aspect of extending battery life in connected devices is to reduce not only the volume of data transmitted to the cloud, but also the frequency and duration of the wireless transmission. Today’s powerful yet energy-efficient microcontrollers (MCUs) make pre-processing a viable option for reducing the volume of raw data transmitted from dozens of sensors across billions of connected devices, but efficient in-situ processing necessitates much greater and faster local memory capacity for storing the algorithms and data to be processed.
Embedded memory can play an important role in meeting energy requirements, as lower power and lower voltage operation, monolithic integration, faster read and write times, non-volatility, and higher capacity are ways that memory technology can help IoT devices achieve greater energy efficiency. For instance, to reduce the frequency of data transmissions designers must make greater use of local data buffering (storage), as batching data for transmission allows the frequency of transmissions to be significantly reduced. Most energy-efficient connected devices integrate sensor hubs and local data logging, and faster memory read and write times enable this data to be stored, processed, and sent more quickly, which reduces the duration of each transmission and optimizes on/off duty cycles with faster wake-up times. On-chip memory enables the use of wide memory buses and lower latencies that break the performance bottleneck between computing cores and storage. Because it is faster than off-chip memories, on-chip memory has a direct impact on energy savings as active processing cycles are reduced and standby or deep-sleep cycles are extended, and monolithic integration of storage memory also offers many cost benefits by eliminating the need for a variety of mechanical connectivity components.
An obvious, if not mandatory choice for lowering the standby current of an IoT device is non-volatile memory (NVM) technology, as by design it can be completely powered down but still retain all of the information it stores.
ReRAM – Power-efficient memory for connected embedded devices
Resistive random-access memory (ReRAM) is a promising technology for developing new high capacity, high performance storage solutions that are both scalable and reliable. Unlike flash-based technology, ReRAM memory is byte addressable and can be architected with small pages that may be independently erased or re-programmed, which drastically simplifies the complexity of the storage controller by removing a large portion of the background memory accesses required for block-oriented data management and garbage collection.
ReRAM cells typically employ a switching material sandwiched between two metallic electrodes that can exhibit different resistance characteristics when a voltage is applied across it, and significant performance gains can be achieved depending on the switching materials and memory cell organization chosen. For example, Crossbar ReRAM technology is based on a simple two-terminal device structure using CMOS-friendly materials and standard CMOS manufacturing processes. It can be easily integrated with CMOS logic circuitry and manufactured using existing CMOS fabs without the need for any special equipment or materials. As it is a low temperature back-end-of-line (BEOL) process integration, multiple layers of ReRAM arrays can be integrated on top of CMOS logic wafers to build system on chips (SoCs) and other chips with large amounts of 3D monolithic embedded RRAM storage.
Compared to flash-based technology, ReRAM achieves visible benefits in terms of lifetime, energy consumption, and read/write latencies (Crossbar ReRAM technology has exhibited 100x lower read latency – on the order of a few hundred nanoseconds – and 20x faster write performance). At the memory cell level, ReRAM improves programming performance and power consumption by achieving a 64pJ/cell program energy, a 20x improvement compared to NAND flash technology. At the system level, having storage memory on-chip reduces energy consumption even further by reducing or eliminating accesses to external memory, which thereby reduces power consumption due to fewer I/O operations. Lower, more predictable latencies also reduce power consumption by shortening the execution time of code fetching or data streaming.
ReRAM technology is also highly reliable and achieves cell behavior that is very stable across a wide temperature range by using a switching mechanism that is based on an electric field. Filamentary nanoparticles and simple CMOS compatible materials like nonconductive amorphous silicon (a-Si) are used as the switching material. When an electric field is applied across the cell, a metallic filament forms across the cell and changes its resistive characteristics.
Bridging the gap between compute and data storage in the Internet of Things
With such capabilities, new radical innovations can be designed in the connected IoT device world through low energy, fast, NVM that can be easily integrated in very large capacities on a single SoC along with logic, analog, and RF components that can operate for years without a battery charge or change. Along with all of its energy-efficiency features, ReRAM has manufacturing process and cost benefits, higher reliability, and the advantage of monolithic integration for single-chip IoT system solutions.
Imagine CES in a few years, firmly in the era of energy-efficient interconnected devices and services improving human productivity. It will be more about data gathering by IoT devices and how to create value from it with accurate analytics, predictive actions, differentiated products, and amazing user experiences. With ReRAM being at the center of this revolution, the gap between computing and data storage will be removed in favor of data-centric computing architectures.