When we think of our smartphones and other computing devices, memory is often not the first feature we see on data sheets. The processor often takes center stage, but memory is the real force behind a device’s ability to get the job done. Flash memory has dominated the memory market of late, but with Moore’s Law it has faced some scaling issues, leading the industry to look elsewhere for memory solutions. Hewlitt Packard’s (HP) hype for “The Machine” has brought the term “memristor” back into the memory spotlight. This technology, also known as Resistive Random Access Memory (RRAM), is being researched and developed to be the next evolution in memory.
Non-volatile memory like flash memory is important to all types of systems because it saves energy by shutting down memory when not in use—especially important in power-constrained embedded systems. But as applications push for faster, higher performance and lower power consumption in smaller packages, memory companies are looking to RRAM’s capabilities to surpass the performance of flash as it approaches its scaling limits. In addition, universities such as Arizona State University are analyzing the advantages and disadvantages of RRAM in various applications.
Comparing RRAM and Flash
The concepts behind RRAM technology are not new — they’ve been around since the 1960s, but have only gained a lot of interest in the past 10 years as the successor to current memory technology. Resistors, capacitors, and inductors are the three basic building blocks of circuits, but the memristor is the theoretical fourth. Memristors are resistors that can remember their history, thus acting as memory, and RRAM is the technology that implements this concept. RRAM devices can maintain a low or high resistance state depending on positive or negative voltages, respectively, which can be read as bits. These states persist when powered down, so it has the potential to be the next non-volatile storage technology.
ASU researchers have been actively developing RRAM technology. Prof. Michael Kozicki is a pioneer in the development of a type of RRAM, the Programmable Metallization Cell (PMC) and its commercial variant, the Conductive Bridge RAM (CBRAM). Professor Kozicki and Associate Professor Hugh Barnaby have also been researching how to make RRAM technology usable in extreme environments such as space, where a combination of low power consumption and non-voltage is essential. Professor Sarma Vrudhula is an active proponent of RRAM technology for new types of computing. Assistant Professor Shimeng Yu has been conducting RRAM research since 2008. RRAM technology is faster (<10 ns) and has a lower programming voltage (<3 V) than current flash memory (<10 µs and >10 V), Yu said.
RRAM is also expected to be more reliable than flash memory, Yu said. Memory reliability is judged in terms of endurance (the number of write cycles before integrity is lost) and retention (the readable age of the data). Compared to RRAM, non-volatile flash has a lower endurance of 10^4 to 10^5 cycles. RRAM can reach 10^6 to 10^12 cycles. Yu said the typical retention standard for nonvolatile memory is 10 years at 85°C, a requirement that can be met by flash memory and potentially by RRAM.
The obstacle to RRAM becoming the successor to flash in the short term is the cost per bit. Flash memory is a very cheap manufacturing technology. Breakthroughs in 3D flash technology have further lowered the cost-per-bit of flash memory, delaying companies like SanDisk’s RRAM roadmap for several years until a cheaper, higher-volume manufacturing strategy can be developed for RRAM devices, Yu said. And the performance gain is not enough to overcome the increased cost of switching to RRAM.
However, the memory market is not the only application for RRAM. Researchers are working on “synaptic” applications, or making computers more like the brain.
Today’s computing architectures work in sequential operations. The CPU fetches data from memory and performs calculations. But this often leads to bottlenecks. The proliferation of data in today’s applications has led to the idea of brain-like ways of processing data in parallel, Barnaby said. In our brain’s neural network, synapses connect active neurons in our brain as we learn. The idea is to use RRAM memory as synapses between artificial neurons in a circuit. This will benefit applications that involve some intelligence, such as image recognition and speech recognition, Barnaby said.
With these exciting developments, it’s an exciting time to use memory technology that could soon steal some processor attention.
Reviewing Editor: Guo Ting