Emerging memory technologies are being investigated as a way to extend information processing as conventional CMOS scaling approaches physical and practical limits. For automation engineers and technical decision-makers, the important question is not simply whether a new memory cell can store data. It is whether the cell, its selector, the fabrication process and the surrounding architecture can work together reliably in a usable system.
The 2015 International Technology Roadmap for Semiconductors frames this challenge around integrating new memory options with CMOS and identifying alternatives for applications that existing memory technologies may not serve as effectively. The roadmap treats these devices as research candidates—not as guaranteed replacements—and emphasizes the scientific, manufacturing and system-level issues that must be resolved.
Why explore emerging memory technologies?
SRAM and flash face scaling limits, while each established memory type involves trade-offs in speed, density, volatility and cost per bit. SRAM occupies significant processor area when used for cache, and increasing that area does not automatically improve net information throughput. Volatile memory also depends on power to retain data, while nonvolatile storage can involve slower access.
Researchers are therefore exploring electrically accessible memories that could combine useful characteristics of different technologies. A high-speed, dense, low-power memory that retains data without power could support new system designs. But achieving those characteristics in a device that can be manufactured and integrated with CMOS remains a substantial challenge.
Storage-class memory (SCM) illustrates the range of requirements. It describes a category intended to combine aspects of solid-state memory, including performance and robustness, with the archival role and low cost per bit associated with conventional hard-disk storage. The roadmap distinguishes faster M-class and slower S-class SCM targets, reflecting that different applications may prioritize different combinations of performance and storage characteristics.
Memory density depends on the selector, too
A memory array is more than a collection of storage cells. Each cell needs a storage element and a way to select it for reading or writing. In resistance-based memories, the storage element may be capable of very small dimensions, but the selector can become the limiting factor for array density.
Planar transistors are commonly used as selectors, but their footprint can constrain how closely cells are packed. Vertical transistors are one proposed route to denser layouts. Another is a two-terminal nonlinear device, such as a diode, used as a separate selector or through nonlinearity inherent in the memory element itself. These approaches must be assessed not only by area, but also by operating current, speed, on/off behavior and compatibility with the array.
This is especially important in crossbar architectures, where many cells share lines. Without effective selection, unintended current paths—often called sneak currents—can interfere with reading or writing. The selector is therefore a system-level part of the memory design, not an optional detail.
Different device mechanisms, different trade-offs
Resistive RAM
Redox-based resistive RAM, or ReRAM, stores information through changes in electrical resistance associated with ion movement and redox reactions in a metal-insulator-metal structure. The roadmap groups reported approaches into several categories, including electrochemical metallization bridge devices and metal-oxide devices with bipolar or unipolar switching. Some are filamentary: a conductive path forms in the material and is subsequently altered during switching. Other approaches use more uniform changes across the device.
These mechanisms create different engineering concerns. Filament formation can be difficult to control consistently, contributing to variability in switching behavior. Some devices require an initial electroforming step, while others face challenges involving switching current, retention, endurance or integration of their materials into CMOS process flows. For array applications, device-to-device variation matters because it affects the ability to distinguish stored states and operate cells predictably.
Polymer-based ReRAM and carbon-based memory approaches broaden the research landscape, but their operating mechanisms and material behavior vary. In many polymer systems, the origin of switching is not fully understood. That makes it difficult to define general material design rules or to compare candidate devices on an equal basis.
Ferroelectric memory
Ferroelectric devices encode information using electrically adjustable polarization. Ferroelectric field-effect transistors (FeFETs) use polarization to shift transistor threshold voltage, allowing a nondestructive read. Ferroelectric tunnel junctions (FTJs) use polarization to alter the resistance of a thin barrier, also supporting nondestructive readout.
Both concepts face trade-offs involving scaling, retention and endurance. In FeFETs, the gate-stack design affects how voltage is distributed across the ferroelectric and insulating layers; that distribution can influence data retention and cycling reliability. For FTJs, scaling the ferroelectric layer must be balanced against maintaining enough read current and controlling variability. Demonstrations at the individual-device level are not, by themselves, enough to establish behavior in large arrays.
Integration and reliability are central design questions
For any emerging memory candidate, cell performance is only part of the assessment. A practical evaluation also asks whether materials can withstand fabrication conditions, whether they are compatible with CMOS processing, and whether reliability issues can be identified early. Heterogeneous integration—the combination of different materials or functions on a CMOS platform—can extend what the platform does, but also introduces process and interface challenges.
Engineers assessing a candidate technology should consider:
- Array behavior: Can selectors address cells while limiting unwanted current paths?
- Consistency: Are switching and resistance states sufficiently controlled across cells?
- Retention and endurance: Do stored states persist, and can the device tolerate repeated operation?
- Process compatibility: Can materials and structures be incorporated without undermining fabrication or reliability?
- System fit: Does the technology suit the intended balance of speed, density, power and nonvolatility?
Beyond the memory cell
The roadmap places emerging memory within a broader effort to sustain information processing beyond conventional CMOS scaling. One nearer-term direction is to integrate specialized devices with CMOS as functional components or accelerators. A longer-term possibility is a new information-processing technology, potentially using different data representations or computational models. Neither direction depends on a memory device alone: interconnect, architecture and the ability to manufacture and integrate the complete system also matter.
For industrial automation, the practical implication is to evaluate emerging memory as part of an application and system architecture, rather than by headline cell dimensions or a single performance metric. The most promising options will need to pair a viable storage mechanism with effective selection, dependable array operation and a credible CMOS integration path. Until those elements are demonstrated together, emerging memory technologies remain candidates for development—not drop-in replacements for established devices.