Access Official Technical PDF

Share Summary:LinkedInEmail

Semiconductor process integration connects materials, device structures and fabrication steps into a working technology. Research presented at the 37th Annual Microelectronic Engineering Conference at RIT illustrates the breadth of that challenge: integrating emerging memories with CMOS, establishing repeatable etch and deposition processes, and tracing defects back through complex manufacturing sequences. For engineers and technical decision-makers, the work underscores a practical point: promising device concepts depend on disciplined process development and measurement at every stage.

Semiconductor process integration for emerging memory

Resistive random-access memory (RRAM) and ferroelectric field-effect transistors (FeFETs) illustrate two approaches to non-volatile memory that rely on different device behaviors. Their development also shows why compatibility with established CMOS processing is an important integration consideration.

In RRAM, a device’s resistance can be changed to represent information. Memristors based on RRAM are also being investigated as artificial synapses for neuromorphic systems, where conductance states can represent synaptic weights. A CMOS-compatible integration strategy developed for a 300 mm wafer platform was described for both front-end-of-line (FEOL) and back-end-of-line (BEOL) configurations. The associated performance challenge is switching variability: inconsistent conductance states can limit the number of usable levels in an analog synapse.

FeFETs use a ferroelectric material in the transistor gate stack. Their two threshold-voltage states can encode binary values, and the difference between those thresholds is known as the memory window. A fabrication process using atomic layer deposition (ALD) of aluminum-doped hafnium oxide was developed at RIT. Initial devices showed ferroelectric behavior and a reported memory window of approximately 150 mV. However, high off-state leakage and a poor on-to-off current ratio pointed to problems with source-and-drain formation. The work also found that threshold-adjustment implantation shifted the transfer curves without degrading the memory window. This combination of promising device behavior and unresolved process issues is typical of early integration efforts: a useful result does not remove the need to diagnose the full transistor process.

Process control: etch, patterning and defect investigation

Etch development is central to defining device geometry, preserving underlying layers and achieving repeatable profiles. A characterization project using a TRION III etcher examined nitride and polysilicon etch behavior, including etch rate, selectivity and profile. The results indicated that power and pressure affected uniformity and profile, while oxygen percentage played a significant role in uniformity. Polysilicon etch rates also differed substantially from previously established Drytek tool data, highlighting why recipes cannot be assumed to transfer unchanged between tools.

Patterning results from a self-aligned double patterning (SADP) project reveal a related integration dependency. The process work qualified coating and deposition steps, including spin-on carbon, bottom antireflective coating, photoresist, silicon nitride and oxide. But angled sidewalls on oxide mandrels prevented nitride spacers from forming properly. The project identified the absence of a hard mask as a likely contributor to the non-anisotropic etch. In practical terms, downstream spacer formation depended on achieving the right profile in an earlier etch step.

Defect analysis can also require separating process interactions rather than changing several parameters at once. In a solar-cell anti-reflective coating investigation, a distinctive “Tiger Stripe” defect pattern helped guide experiments: wafers were rotated in different orientations at suspect processing steps to identify where the defect began. The investigation traced the issue to residual chemical drying on the wafer during a rinse operation, which became visible after vacuum processing. Adjusting the rinse parameters eliminated the defect. The broader lesson is that visual patterns can serve as clues, while controlled experiments help pinpoint the originating step.

Materials and device structures expand the process window

Other projects explored how material properties can enable new functions while creating distinct fabrication and characterization requirements. Silicon-nitride optical waveguides, for example, were fabricated with loopback and ring-resonator designs at two of the tested film thicknesses. Grating couplers resolved at pitches from 900 nm to 1,500 nm, while pitches of 800 nm and below were overexposed and did not develop. Optical testing could not be completed because the test setup was unavailable, so simulations were used to examine expected transmission conditions. Separating fabricated results from simulated results is essential when assessing process maturity.

Work on two-dimensional materials included tape transfer of molybdenum disulfide (MoS2), with Raman spectroscopy used to examine layer-dependent characteristics. Bilayer transfer was demonstrated, while monolayer transfer was considered possible but could not be confirmed beyond visual inspection with the available tools. The next step identified in the project was to fabricate devices and conduct electrical testing.

Device fabrication studies also covered AlGaN/GaN high-electron-mobility transistors (HEMTs), indium-gallium-zinc oxide thin-film transistors and N-polar GaN HEMTs. The initial AlGaN/GaN fabrication effort established a baseline despite limited gate control, substantial gate leakage and Schottky contacts in device regions. In a separate N-polar GaN study, a deep-recess structure with a GaN cap in the access regions was associated with reduced DC-to-RF dispersion and improved access-region conductivity. The report described high output-power density and power-added efficiency over operating frequencies from 30 to 94 GHz. These examples show how device architecture, material orientation and process execution interact; they should not be treated as interchangeable process recipes.

What the findings mean for manufacturing teams

Across memory, photonics, power devices and patterning, the projects point to a consistent approach for moving emerging technologies forward:

  • Qualify individual process steps. Measure film thickness, etch rate, uniformity and selectivity rather than relying on nominal settings.
  • Connect upstream profiles to downstream results. An etch profile can determine whether later spacer or device structures form correctly.
  • Use electrical and materials characterization together. I-V behavior, microscopy and spectroscopy answer different questions about device function and process quality.
  • Design experiments to isolate causes. Controlled changes can distinguish a specific process failure from interactions among multiple steps.
  • State evidence limits clearly. Fabrication, simulation and completed electrical or optical testing are distinct levels of validation.

For industrial automation and semiconductor manufacturing teams, the practical implication is that process integration is a system-level engineering task. Device concepts become useful only when materials, equipment, recipes, measurement methods and downstream steps work together reproducibly. Early results—whether a measurable memory window, a resolved waveguide pattern or a traced defect—can establish a basis for improvement, but careful characterization remains essential before drawing conclusions about production readiness.

Access Official Technical PDF

About The Author