IIoT encompasses interconnected electronic systems pivotal to agriculture, energy production, automotive manufacturing, and other complex industrial sectors. These systems’ advanced functionality – high-speed connectivity, AI-driven decision-making, and continuous monitoring for optimisation and maintenance – generates a constant influx of data from sensors and localised processing units.
A means of reliable, secure data capture is therefore central to IIoT efficacy, which places stringent demands on the memory design of control systems within the IIoT framework. Programmable logic controllers (PLCs), crucial for automation in factories and control centres, require robust memory solutions to manage factory line configurations, sensor data, and machine health monitoring. Magnetic random access memory (MRAM) has emerged as a key technology in this context.
What is MRAM?
Controlling actuators, sensors and other types of input/output (IO) devices in factory automation applications, PLCs provide high flexibility for factory processes due to their programmability and scalability for handling from hundreds to thousands of IO devices. Making these systems rugged involves ensuring operation over a wide temperature range and protecting critical system data in the event of power loss. Because customer line configurations and system state information must be protected and instantly retrievable under any circumstances, PLC design requires a non-volatile memory (NVM), which maintains data when power is lost or interrupted.
However, the memory must not only be non-volatile; it must also be able to withstand constant updating and retain data under high temperatures for extended periods of time. MRAM is the memory of choice for PLC manufacturers due to its unique combination of attributes that meet the need for data protection, fast access times, high write cycle endurance and long data retention – all in one memory device.
More specifically, MRAM’s key features include:
- At least ten years of data retention at temperatures up to 105°C.
- Near-unlimited writes over the life of the customer system.
- Symmetric read and write speeds and access times, with byte-level addressability.
- 400 Mbytes of read and write bandwidth for the most recent industrial MRAM.
- Industrial operating temperature range of -40C to 105C.
Why MRAM?
PLCs have employed MRAM for more than ten years. Its advantages over alternative memory devices are driving an increase in its adoption. MRAMs offer design-in flexibility because they operate with standard bus interfaces, both parallel and serial, making them compatible with SRAM and serial NOR devices that have the same interfaces. Its inherent non-volatility and extremely high endurance also facilitates MRAM’s design-in because there is no need for external capacitors or batteries to sustain power to the device in the event of power loss. This allows for greatly simplified system power-loss-protection design.
MRAM’s high endurance means there is no need to manage the frequency or locations of writes in the memory device, as is required in flash-based NVMs. Moreover, new MRAM models use the relatively new xSPI JEDEC serial interface standard, further extending ease of use and standardisation for system designers.
Other NVMs have limitations that MRAM does not possess. Below are a few examples.
Nonvolatile static RAM (nvSRAM) is not scaling to higher densities because of a highly complex memory storage cell that relies on both SRAM bit cells, which are backed up by Flash-like cells.
NOR Flash is good for code storage that does not need much updating, but it cannot handle the volume of writes needed for system monitoring.
Battery-backed SRAM (BB-SRAM) offers good read and write speeds, but requiring a battery to retain data in the SRAM when power is lost makes it less desirable for PLCs. Batteries are often unreliable and eventually have to be replaced, making a BB-SRAM system susceptible to downtime or field replacement issues.
A leg up for Edge AI
The use of AI in process automation is a new trend, and systems are in development that adopt AI inferencing and machine learning. AI processing is often accomplished with a dedicated accelerator that will perform inferencing within a system, which then provides interpretation and optimisation of process conditions, predictive maintenance, or camera recognition. These systems are often at the Edge or far Edge of the network, and there is an advantage in performing the acceleration function locally rather than in Cloud-based systems.
Edge AI processing in control systems enables rapid, localised optimisation, reducing reliance on Cloud-based data centres. Its flexibility, nonvolatility, and endurance for continuous writing make MRAM an ideal memory solution for the dynamic needs of IIoT. It facilitates local model storage and quick updates, with instant-on capability and very fast writes to speed over-the-air updates and system feature additions – all capabilities essential for adapting to evolving customer demands. This technological advancement allows for versatile deployment within the memory hierarchy, offering system designers distinct options to cater to varied memory workloads in the industrial domain.
Versatility for multiple Industrial IoT applications
MRAM is highly versatile, allowing it to be optimised for specific system memory workloads. As a persistent data memory that stores parametric, configuration, or system logs, MRAM is optimised for write-cycle endurance of 1e14 cycles as well as very fast write speed. In the case of a unified code and data storage memory, the MRAM can be optimised for higher capacity and tradeoff of write speed and cycles. MRAM as a discrete memory component is not typically customized, but the design of the memory can be optimised for the workload and maintain standard interfaces for the system.
Everspin has seen MRAM’s adoption for numerous industrial use cases. Just a few examples include the Siemens Simatic MP 277 and MP 377 Multipanel Machine Control Systems, Advantech PCM-23 Series Industrial Computing, Emerson Network Power’s MVME7100 Industrial Computer Board, and Schneider Electric Modicon M580 Automation Controller.
Beyond process control, MRAM is deployed in smart grid controllers for utilities, in solar power inverters, building lighting controllers, electric train controllers and ticket machines and many other embedded applications within the infrastructure of smart cities. The future is bright for the expansion of MRAM in the IIoT as AI places more demands on memory systems.
Joe O’Hare is Director of Marketing at Everspin Technologies.