Access Official Technical PDF

Share Summary:LinkedInEmail

Logic and memory devices sit at the heart of semiconductor systems, but continued scaling is making their design increasingly difficult. Density, performance, and cost remain important, while dynamic and static power have become major device-level constraints. For industrial automation engineers and technical decision-makers, these developments matter because they influence the computing hardware available for control, sensing, data processing, and embedded systems.

Several research directions aim to improve transistor efficiency, address parasitic losses, and develop memory cells that can meet demanding requirements for density, speed, power, and reliability. Each approach brings distinct trade-offs; progress depends not only on demonstrating a promising physical effect, but also on achieving the full set of operating and integration requirements.

Why logic and memory devices are difficult to scale

In conventional charge-based transistors, reducing supply voltage without sacrificing drive current is a central objective. One route is to use channel materials with higher carrier mobility. Another is to seek a sharper turn-on response, which could allow a transistor to operate at a lower voltage while maintaining the required distinction between on and off states.

Scaling also exposes limitations within the transistor itself. In particular, internal series resistance can substantially reduce device current, and the challenge is expected to become more severe as dimensions shrink. Improving contacts, junctions, and gate structures is therefore relevant across technology generations—not just to emerging device concepts.

New transistor approaches for lower-power logic

High-mobility channels and tunneling devices

Nanowire MOSFETs represent a potential progression from planar silicon-on-insulator MOSFETs and FinFETs, while retaining silicon channels. Research also explores alternative channel materials, including III-V compounds, germanium-based materials, graphene, carbon nanotubes, and two-dimensional materials such as transition metal dichalcogenides. Higher mobility may support stronger current or lower operating voltage, but a candidate material must meet key requirements together, including on-current, off-current, and low-voltage operation. Compatibility between n-channel and p-channel devices is another important consideration.

Tunneling transistors pursue a different route to lower-voltage operation. Conventional MOSFETs face a fundamental subthreshold-slope limit of 60 mV per decade. This constrains how far supply voltage can fall while preserving a needed on-current-to-off-current ratio. A tunnel field-effect transistor, or TFET, uses a gate-controlled p-n junction to produce current through tunneling and can, in principle, achieve a steeper subthreshold slope. Other concepts explore band alignment, bandgap tuning, or tunneling between material layers separated by an insulator.

Other operating principles

Not all alternative logic devices rely on tunneling. Negative-capacitance FETs incorporate a ferroelectric layer in the gate dielectric and seek to use internal voltage gain. Other investigated concepts include devices based on metal-insulator transitions, piezoelectric effects, impact ionization, and phase transitions. Functional devices are also of interest when a single structure can reproduce a logic operation otherwise implemented with a network of multiple devices.

Spin- and magnetism-based devices use a state variable other than charge or voltage. They may offer a way to avoid resistive power dissipation, but practical challenges include propagating spin information, performing logic operations with it, and converting spin states back into electrical signals.

Emerging nonvolatile memory and array design

Memory technologies serve different roles and come with different performance and structural trade-offs. Traditional charge-storage nonvolatile memories use floating-gate or charge-trapping transistors. Their tunnel oxide must support long retention, and the thickness has reached a point where further scaling is difficult. Three-dimensional structures can use vertical space, while interest has also expanded to non-charge-storage cells, often with two terminals.

Key considerations for these emerging nonvolatile memories include write and erase power, density, multi-bit capability, endurance, reliability, and latency. Technologies under development include:

  • MRAM: Magnetoresistive RAM, including spin-transfer-torque RAM. Research targets include free-layer materials and structures that reduce switching current, as well as reference layers and tunnel barriers that support high magnetoresistance.
  • RRAM: Resistive RAM, including approaches based on oxygen-ion or vacancy movement in an oxide, and conductive-bridging RAM, which forms a conductive metal filament.
  • PCRAM: Phase-change RAM.
  • FeRAM: Memory based on ferroelectric materials and effects.

Analog nonvolatile memories offer a continuum of equilibrium states rather than only discrete states. Their potential use in neuromorphic computing makes controllability and reliability important research issues. Symmetric resistance changes in response to pulsed inputs are a particularly relevant goal.

Why selector devices matter

Some two-terminal nonvolatile memory cells can conduct too readily at low bias in their low-resistance state. In an array, this can create unwanted leakage paths. A selector placed in series with each memory cell can help control current, and a two-terminal selector is attractive because it can conserve cell area. However, meeting the necessary selection requirements has not been fully demonstrated. An ideal direction is a memory cell with built-in selection behavior—one that has negligible current at low bias in its low-resistance state.

Scaling also challenges conventional six-transistor SRAM and one-transistor, one-capacitor DRAM cells. Lower supply voltage can create noise-margin difficulties, while area reduction remains a persistent concern. New materials or simpler structures that provide the same memory functions with fewer device components could help address both issues.

Integration, modeling, and reliability

Device advances must be evaluated in the context of manufacturing and system integration. Embedded logic or memory may need to fit within three-dimensional, back-end, or interconnect processes. Simulation and modeling can support the study of materials, process steps, device behavior, and quantum effects in nanoscale geometries. More capable and computationally efficient methods are important for assessing concepts before they can be fully realized.

Metrology is also essential: new methods and platforms are needed to extract memory performance and relevant material characteristics. Reliability cuts across both logic and memory, so it should be assessed in relation to the specific device and its operating mechanisms.

Conclusion: Evaluate devices as complete systems

The next steps in logic and memory device development are not defined by one material or architecture alone. Lower-power transistors must deliver practical on- and off-state behavior; memory cells must balance density, speed, energy, endurance, and reliability; and selectors, contacts, integration, and measurement methods must support operation at the array and process level. For automation technology decision-makers, the practical implication is to evaluate emerging semiconductor concepts against complete application and manufacturing requirements—not a single attractive device metric.

Access Official Technical PDF

About The Author