The memristor memory part is what we hope has caught your attention now. Memristor memory is a non-volatile memory that uses memristors or electrical components that regulate the flow of current in a circuit and remembers the amount of charge that has previously flowed through it.
If Intel bets the future of SSDs on 3D XPoint, HPE has its hopes for memristor memory. HPE and SanDisk have now partnered to develop a new storage class memory technology, which will combine the best of HPE’s memristor project and SanDisk’s non-volatile resistive random access memory (ReRAM) technology. ReRAM, too, is a memristor based technology.
Rough sketches show that the design will involve tiny conductive filaments crisscrossing and connecting silicon layers to represent a bit of data. The resulting non-volatile memory technology is expected to be up to thousand times faster and thousand times more durable than current flash technology. (This is exactly what Intel says about 3D XPoint, so we can well assume that this is how SSDs of the future will be!) It will be byte-addressable like DRAM and will allow systems to employ several terabytes of storage class memory cache per server node for large-scale applications. The resulting technology will be used not only for The Machine but also to make other commercial and enterprise products.
Other companies are also working towards commercialising ReRAM technology by this year. One such is Silicon Valley start-up Crossbar, who hopes that some of their 3D Restive RAM products will be used as memory in wearable devices this year, and in high-density SSDs sometime next year or later.
Looking further into the future
There is a lot of exciting research happening at universities, which hint at what is in store for SSDs well into the future. Here are some examples:
Tantalum based memory. One such research was revealed by scientists of Rice University last year. A team of chemists led by James Tour discovered that applying a voltage to a 250-nanometre-thick sandwich of graphene, tantalum, nano-porous tantalum oxide (an insulator) and platinum creates addressable bits where the layers meet. Control voltages can be used to shift oxygen ions and vacancies, to switch the bits between ones and zeroes. The team used this to create a solid-state memory technology that allows for high-density 162-gigabit non-volatile storage. The new devices require only two electrodes per circuit, which is simpler than current-generation flash memories.
According to Tour’s comments in the university’s press reports, “This tantalum memory is based on two-terminal systems, so it’s all set for 3D memory stacks… And it doesn’t even need diodes or selectors, making it one of the easiest ultra-dense memories to construct. This will be a real competitor for the growing memory demands in high-definition video storage and server arrays.”
Further, the tantalum oxide memories can be fabricated at room temperature. The control voltage for write/rewrite is adjustable, allowing a wide range of switching characteristics. The main challenges in the way of commercialising this memory technology are the fabrication of a crossbar device dense enough to address individual bits and a way to control the size of the nano-pores.
BlueDBM. Although not an SSD technology by itself, BlueDBM, an ongoing project at the Massachusetts Institute of Technology has the potential to improve the usage of upcoming flash based storage technologies. A system architecture to accelerate Big Data analytics, BlueDBM comprises a large distributed flash based storage with in-store processing capability and a low-latency high-throughput inter-controller networks.
In some Big Data scenarios, performance is affected by the capacity of fast local DRAM, and in cluster systems with more RAM, the network stack ends up being a bottleneck. BlueDBM attempts to overcome these issues by providing extremely fast access to a scalable network of flash based storage devices, and to provide a platform for application-specific hardware accelerators on the data path on each of the storage devices.
According to the project overview, “Each BlueDBM node consists of flash storage coupled with a field-programmable gate array (FPGA), and is plugged into a host system’s PCIe port. Each node is connected to up to eight other BlueDBM nodes over a high-speed serial link capable of 10-gigabit bandwidth at 0.5µs latency. By default, the FPGA includes platform functions such as flash, network and on-board DRAM management and exposes a high-level abstraction.”
Photonic memory. In a paper published in nature.com last year, a group of scientists from English and German universities proposed an integrated all-photonic, non-volatile multi-level memory. Photonic data storage can greatly improve the performance of computing systems by reducing the latencies associated with electrical memories and potentially eliminating optoelectronic conversions. While scientists have experimented with photonic memory earlier, these attempts mostly resulted in volatile memory.
The recent breakthrough uses phase-change materials to achieve a robust, non-volatile, all-photonic memory. In simple terms, this solution uses waveguide technology to move light from lasers across a germanium, tellurium and antimony alloy nano-coating. The structure of the alloy is altered in predictable patterns when hit by light from a high-intensity laser. A low-power laser is used to read the patterns and translate it as data. By using optical near-field effects, the team has achieved bit storage of up to eight levels in a single device that readily switches between intermediate states. According to the paper, the on-chip memory cells feature single-shot readout and switching energies as low as 13.4pJ at speeds approaching 1GHz.
Before they think of commercialisation, the team has to figure out how to reduce the size of their prototype, which is much larger than current-generation memory chips! However, in terms of performance, photonic memory can easily outperform SSDs, so it is a worthy pursuit.
SSDs cannot be monarchs
Speaking on the future of SSDs at last year’s SNIA Storage Developer Conference in Santa Clara, California, Jim Handy from Objective Analysis said, “Future memory and storage systems will include everything: DRAM, non-volatile memory, NAND and hybrid hard drives. One technology will not kill off the others.”
Jim added, “The result of all the changes in technologies and architectures is a new component: storage-class memory. This memory will combine the benefits of a solid-state memory with DRAM performance and robustness and archival capabilities and low cost of hard-disc magnetic storage. The persistent memories will disrupt the entrenched thoughts on possible and available latency budgets for storage… The computer of tomorrow will have fixed DRAM size made of stacked packaging of DRAM dice, and upgradeable non-volatile memory that will be the equivalent of DIMMs. The storage system will include both flash and disk with the flash on PCIe of its own bus. The magnetic drives will continue to exist for mass storage as there is no foreseeable price crossover for high-density, long-term storage. Finally, the storage class memory software will eventually get to the point where is will contribute to overall system performance.”
So it is clear that, while we will see more SSDs around us in the coming years, an all-flash era might be quite far away.
Like this article? You may also like other Tech Focus articles
Janani Gopalakrishnan Vikram is a technically-qualified freelance writer, editor and hands-on mom based in Chennai