IoT and embedded applications require ever more RAM with higher bandwidths, smaller form factors and lower power consumption. For developers, this presents the question of what an ideal memory module should look like for such applications. Chen Grace Wang, Product Manager at Rutronik, and Wesley Kwong, Business Development Manager at AP Memory explains.
The user experience benchmark for IoT and embedded applications continues to get higher and higher, which demands more RAM with higher bandwidths, small form factors, lower power consumption and thus also less power loss while also keeping the component costs the same or lower. This is especially true of applications that use Artificial Intelligence (AI) and/or Machine Learning (ML).
SRAM (static RAM) is still the RAM solution that offers the highest speeds and lowest latency and is closest to the processor, but it does have some drawbacks. The regular 6T-SRAM layout topology has not been shrunk to the same proportion as the process nodes.
The power loss of embedded SRAM also increases as the CPU consumes more power. This means that it is becoming increasingly difficult to meet the requirements of the latest IoT applications using embedded SRAM due to the limitations in terms of power consumption and their increasing RAM requirements.
External SRAM modules also require a high number of transistors, which increases memory costs. As a result, it is almost impossible to meet the limited form factor requirements.
External DRAM modules (dynamic RAM) still offer considerable cost advantages over SRAM. With a single transistor and capacitor, they offer comparable performance, allowing for a much denser array.
For applications that are persistently or usually attached to a single power supply unit, external DRAM modules may be an acceptable solution. However, they do have a large number of pins, and their update requirements and the ever-increasing complexity of routing mean that they are complicated to integrate.
Older SDRAM modules (synchronous DRAM modules) with low densities are designed for older process nodes and their size makes them basically unsuitable for compact, energy-efficient systems.
This means that a RAM alternative is needed that offers high performance at lower cost and with lower power consumption, while also meeting the growing requirements of a complete IoT user experience.
IoT RAM combines benefits of DRAM and SRAM
IoT RAM is based on pseudo-static RAM technology (PSRAM). It combines the advantages of DRAM – a small surface area, low product costs of up to a tenth of SRAM and a density ten times higher than that of SRAM – with those of SRAM, namely high speed, low latency and ease of control. Internally, PSRAM uses DRAM cells, which consist of just one transistor and one capacitor, but behave like ordinary SRAM and the conventional, relatively simple SRAM interfaces.
IoT RAM also offers flash-SPI interfaces with low pin counts that are used by many MCUs and FPGAs. The low-cost IoT RAM solutions from AP Memory are compatible with the SPI interfaces of most MCUs, SoCs and FPGAs, including Quad-SPI (QSPI) and Octal-SPI (OSPI).
System-in-package (SiP) versions of IoT RAM are suitable for any situation where SoCs require more memory than is possible with the internal SRAM. SiP options, especially those using ‘known good dies’ (KGD), provide all of the aforementioned benefits thanks to the higher system memory, making them ‘more than Moore’.
The low latency of IoT RAM allows very fast wake-up from modes with very low power consumption, immediate wake-up from standby, and fast switch-on times. IoT RAM operates with very low power consumption, usually 0.15 to 0.5µA/mbit depending on the memory density.
Internal refresh
Looking at the example MCU diagram (Figure 2), the spaces for RAM and static memory are continuously growing. If DRAM is used for this, this increases the power consumption of the system, and also requires the integration of a refresh controller.
IoT RAM requires no controller as the entire refresh logic for the DRAM cells – unnoticed by the user – is handled internally. This reduces the complexity of the interfaces and the validation costs that this entails. Older MCU-based systems that still use SDRAM benefit from IoT RAM thanks to lower power consumption and the simplified interfaces (see table).
Fluid video playback in Edge Computing
Looking at an application using frame buffering, it becomes apparent how external RAM enables superior user experiences. The system does not need to access slower non-volatile memory as often for read/write activities, which improves the system performance as a whole. This is shown by the Coremark test suite. The user benefits from lower latency, more fluid video playback and more reliable recording.
IoT RAM solutions from AP Memory already work seamlessly with many existing MCUs, SoCs and FPGAs in IoT/embedded devices where high performance, low cost and fast reaction capabilities are needed. To this end, AP Memory maintains close partnerships with a growing number of MCU, SoC and FPGA providers.
IoT RAM solutions offer simplified signal protocols (QSPI, OPI, ADMUX) and packaging options (KGD, WLCSP, SOP, USON, BGA) for volatile memory in IoT and edge computing products.
Rutronik offers a large selection of both IoT RAM and PSRAM solutions from AP Memory with a variety of memory densities to meet a range of performance and bandwidth requirements.
