Digital technologies and data storage capacities are growing exponentially with each passing year. As the amount of data stored and exchanged globally multiplies, traditional memory technologies like hard disk drives and NAND flash storage are facing limitations in performance, capacity, and scalability. This is pushing researchers and companies to explore innovative next generation memory technologies that could replace or complement existing solutions in the years to come. Let’s take a look at some of the most promising candidates for next gen memory and how they may reshape the data landscape.
Emerging Non-Volatile Memory Technologies
Non-volatile memory technologies offer data persistence even when power is switched off, unlike volatile dynamic RAM. Some of the most notable emerging non-volatile technologies include:
Resistive Random-Access Memory (RRAM): RRAM utilizes the resistance variation in transition metal oxides for data storage. It provides superior performance compared to NAND with near-DRAM speeds, good endurance and energy efficiency. Commercial RRAM products could hit the market as early as 2023 to take on enterprise SSDs and storage-class memory roles. Major players developing RRAM include Intel, Micron, Crossbar and Adesto.
Phase-Change Memory (PCM): PCM utilizes the unique resistance properties of chalcogenide glass to reversibly switch between crystalline and amorphous states representing digital 1s and 0s. Intel and IBM have demonstrated PCM prototypes at the consumer level with significantly faster write/read times than NAND. PCM is commonly seen as the frontrunner for universal memory as its performance is comparable to DRAM but with non-volatility.
Ferroelectric RAM (FRAM): FRAM stores information based on the retention of ferroelectric polarization in crystalline materials like lead zirconate titanate ceramic. FRAM provides unlimited read/write endurance and almost instantaneous access times. It could vie for embedded applications where guaranteed data retention even during power loss is crucial. Fujitsu is a leader in FRAM R&D and production.
Magnetoresistive RAM (MRAM): MRAM uses magnetic elements instead of electric charge to store data. Early MRAM variants faced challenges but newer designs like spin-transfer torque MRAM have shown commercial potential, with Everspin shipping discrete MRAM chips since 2006. MRAM strengths are unlimited endurance and near-instant reads/writes with no wear-out, making it ideal for enterprise storage and automotive/industrial use cases.
The Rise of 3D Memory Architectures
To further maximize density and performance within chip area constraints, companies are exploring 3D stacking of memory components. This allows for much faster data access by reducing bit-to-bit communication distances dramatically. Some notable 3D memory technologies include:
3D XPoint: Developed by Intel and Micron, 3D XPoint memory blends aspects of phase change and ST-MRAM. It provides breakthrough densities at 1000x the performance and endurance of NAND, enabling a new class of “optane” SSDs and storage-class memory products that sit between DRAM and NAND. Optane DIMMs for servers and Intel’s Optane Memory for PCs have already arrived.
Hybrid Memory Cube (HMC): HMC stacks DRAM dies on a logic layer in a TSV-connected cube structure. This achieves up to 10x higher bandwidth than conventional memory at lower power. Companies like Micron and Samsung have shipped HMC modules for enterprise applications. The latest version, HMC 3, promises 224GBps bandwidth.
High Bandwidth Memory (HBM): Used mainly in GPUs initially, HBM vertically stacks DRAM dies connected through through-silicon vias (TSV). This boosts bandwidth to over 512GBps while using 30-50% less space than traditional graphics memory. Players like AMD, Intel and Nvidia rely on HBM for discrete graphics and accelerators requiring massive data throughput.
The Future is Cross-Point Memory
Looking ahead, cross-point or crossbar memory architectures are envisioned as the ultimate path to higher densities and lower costs compared to today’s NAND and even 3D memory stacks. In cross-point designs, memory cells are individually addressable where word-lines intersect bit-lines in a densely packed 2D lattice. This enables terabits of memory to be built within a silicon die just millimeters across.
Phase change and RRAM are ideal technologies for cross-point implementation due to their small switching volumes and high scalability. Companies working on cross-point prototypes based on these memory types include Adesto, Crossbar, HP, Intel and Micron. Going forward, we may see commercial availability of universal non-volatile memory at storage densities in terabits/cm2 within this decade, rendering traditional memory storage obsolete. The data landscape 10 years from now is sure to be unrecognizable from what we know today.
In summary, next generation memory technologies promise to revolutionize how we store, process and share massive amounts of data globally. They herald an era of instant access to exabytes of information from small ubiquitous devices rather than centralized servers. With extensive R&D underway, many of the discussed concepts are expected to scale up and enter commercialization over the next 5-10 years. The resultant compute and data processing capabilities will likely accelerate innovation across industries. Exciting times indeed for the memory technology field!
*Note:
- Source: Coherent Market Insights, Public sources, Desk research
- We have leveraged AI tools to mine information and compile it
