Two-dimensional numerical device simulations have demonstrated that extending the heavily doped source region of a silicon-on-insulator metal–semiconductor field-effect transistor significantly improves high-frequency analog amplification without degrading the device’s electrical breakdown voltage. The numerical investigation separates the distinct physical mechanisms governing source-side series resistance and gate–drain breakdown by comparing four structures of equal gate length. By decoupling source extension from gate repositioning, the computational study reveals that shortening the ungated channel gap on the source side boosts transconductance and microwave power gain while maintaining voltage-handling capability. The expected engineering benefit of these findings is a moderate-voltage, high-frequency device for millimeter-wave gain stages that does not need extra material layers.
Silicon-on-insulator metal–semiconductor field-effect transistors, commonly designated as SOI-MESFETs, utilize a metallic Schottky contact deposited directly onto a thin silicon film to control current flow through an underlying conductive channel. Unlike traditional metal-oxide-semiconductor field-effect transistors, the MESFET architecture operates without an insulating gate oxide layer and does not require high-temperature implant activation annealing. Fabricated on a silicon-on-insulator substrate in a standard CMOS line, such devices can be used for power, RF, and extreme-environment circuits. However, practical deployment has long been constrained by a persistent engineering trade-off between breakdown voltage and operating speed. In conventional geometries, widening the drift region between the gate and drain enhances voltage tolerance but introduces parasitic access resistance and inter-electrode capacitance that degrade the cut-off frequency, maximum oscillation frequency, and transconductance.
Previous efforts across the semiconductor physics literature have addressed this performance bottleneck primarily by engineering electric field distributions across the drift space. Documented techniques include introducing localized oxide and metal inclusions, embedding buried air gaps, inserting silicon carbide channel inserts, and implementing reduced-surface-field doping profiles. Earlier investigations by the current authors incorporated buried metallic layers and shifted the gate electrode closer to the source contact, raising breakdown voltages from 16 volts to 27 volts while enhancing transconductance. Nevertheless, moving the gate electrode simultaneously alters both the gate–source and gate–drain spacings, thereby masking the isolated influence of the source contact geometry. Local source modifications had been documented in split-source tunnel FETs and pocket-engineered nanowire transistors, but the specific impact of lengthening the heavily doped source diffusion in an SOI-MESFET had not previously been decoupled from gate displacement.
The simulated platform comprises a zero-point-two-micrometer n-type silicon channel doped at ten to the seventeenth power per cubic centimeter, situated atop a zero-point-four-micrometer buried oxide layer and a zero-point-one-micrometer p-type silicon handling substrate. The drain comprises a heavily doped n-type region extending zero point three micrometers with a donor density of ten to the twentieth power per cubic centimeter, yielding an active ionized electron concentration of approximately one point five times ten to the nineteenth power per cubic centimeter. The Schottky gate contact exhibits a work function of five point one electron-volts and an effective barrier height of zero point nine three electron-volts.
The benchmark device, designated Structure C, embodies a conventional layout with a centered gate electrode establishing symmetric zero-point-six-micrometer separations between the gate and the source and drain diffusions. Structure B implements source-extension engineering alone: the heavily doped source extends zero point eight five micrometers across the active area, bringing its boundary within zero point zero five micrometers of the centered gate while leaving the zero-point-six-micrometer gate–drain drift spacing completely intact. Structure A evaluates asymmetric gate displacement alone by shifting the gate toward the drain to create a long zero-point-nine-five-micrometer gate–source separation and a shortened zero-point-two-five-micrometer gate–drain drift zone while retaining a short zero-point-three-micrometer source diffusion. Structure LS, representing the long-source configuration, combines both features by expanding the source diffusion to one point two micrometers and positioning the gate at the shifted coordinates, thereby maintaining a minimal zero-point-zero-five-micrometer gate–source separation alongside the narrow zero-point-two-five-micrometer gate–drain spacing.
Simulations performed using the Silvaco ATLAS technological computer-aided design software incorporated physical transport models for Shockley–Read–Hall recombination, Auger recombination, standard band-to-band tunneling, concentration-dependent mobility, and parallel-field velocity saturation with a saturation velocity of ten to the seventh power centimeters per second. The breakdown behavior was evaluated using the Selberherr impact ionization model. The small-signal admittance matrix was extracted across a wide frequency spectrum spanning ten hertz to one hundred and seventy-four gigahertz under identical bias conditions of zero point five volts gate bias and six volts drain bias, with additional evaluations conducted at three volts drain bias to distinguish normal passive small-signal transport from avalanche-induced conduction anomalies.
The comparative numerical results demonstrated that source extension alone, as realized in Structure B, eliminates a significant fraction of the parasitic series resistance inherent to the ungated source segment. In Structure C, this resistive segment degenerates the intrinsic transconductance. Extending the highly conductive source diffusion in Structure B lowered the linear-region on-resistance from seven point four seven kilohm-micrometers to four point nine six kilohm-micrometers. At a drain bias of five volts and a gate bias of zero point five volts, the on-state drain current increased by twenty-nine percent, rising from one hundred and fifty-five milliamperes per millimeter in Structure C to two hundred milliamperes per millimeter in Structure B. Simultaneously, the small-signal transconductance measured at one gigahertz surged by sixty-seven percent, advancing from fifty-six millisiemens per millimeter to ninety-four millisiemens per millimeter.
Crucially, Structure B preserved the full breakdown voltage of the conventional baseline. Under off-state evaluation at a gate bias of minus three volts, both Structure C and Structure B exhibited an identical solver-limit breakdown voltage of seventeen point six volts, which corresponded to thirteen point two volts and thirteen point one volts under a strict drain-current threshold criterion of one hundred milliamperes per millimeter. This quantitative invariance confirmed the primary hypothesis that off-state breakdown is dictated entirely by electric field crowding within the gate–drain drift zone rather than the physical length or proximity of the source diffusion. Parametric scans across various gate-to-source spacings verified that breakdown voltage remained constant at nine point seven volts across separations ranging from zero point fifteen to zero point ninety-five micrometers when the gate-drain spacing was held fixed at zero point twenty-five micrometers.
RF small-signal characterization highlighted important dynamic trade-offs associated with source proximity. Although the sixty-seven percent transconductance enhancement in Structure B was substantial, the cut-off frequency did not increase, registering twenty-four point three gigahertz compared to twenty-five point four gigahertz for the conventional device. Detailed admittance analysis revealed that moving the conductive source contact within fifty nanometers of the gate increased the total gate capacitance by sixty-five percent, rising from three point seventy-five times ten to the minus sixteen farads per micrometer to six point nineteen times ten to the minus sixteen farads per micrometer due to lateral electric fringing fields. However, the maximum oscillation frequency at a reduced drain bias of three volts improved by forty-one percent, climbing from seventy gigahertz in Structure C to ninety-nine gigahertz in Structure B, driven by the sharp reduction in source series resistance.
The composite Structure LS demonstrated dramatic high-frequency gains at the expense of lower voltage tolerance. Because its gate-to-drain spacing was compressed to zero point twenty-five micrometers, its breakdown voltage dropped by forty-five percent to nine point seven volts. In exchange, the combined architecture achieved an exceptional transconductance of one hundred and thirteen millisiemens per millimeter and a cut-off frequency of thirty-one point three gigahertz at six volts drain bias. Mason’s unilateral gain at one hundred gigahertz reached nine point three decibels, exceeding the conventional structure by ten point four decibels. At a three-volt drain bias, where small-signal admittance remained unconditionally passive, Structure LS achieved an indicative maximum oscillation frequency of one hundred and sixty-three gigahertz, compared to seventy gigahertz for the conventional geometry.
The authors also examined fabrication tolerances by modeling a structural variant, designated LSe, wherein the metallic contact metallization covered only the initial zero point three micrometers of the one-point-two-micrometer extended source diffusion. Leaving the remainder of the heavily doped silicon extension uncovered shifted the metallic edge from zero point zero five micrometers to zero point ninety-five micrometers away from the gate. This modification diminished gate-to-metal fringing capacitance by eight point five percent, resulting in minor performance reductions of only three point two percent in drain current, ten point three percent in transconductance, and one point eight percent in cut-off frequency. This demonstrates that the source-extension mechanism remains highly effective even without sub-optical lithographic metal alignment, significantly relaxing manufacturing constraints for practical foundry implementation.
Despite these encouraging numerical findings, the authors emphasized critical physical and methodological limitations that require experimental validation. The simulations relied on default, uncalibrated Selberherr impact-ionization parameters and an idealized drift–diffusion transport model that neglects non-local velocity overshoot and hot-carrier heating phenomena prevalent above one hundred gigahertz. Furthermore, numerical mesh sensitivity tests demonstrated that peak electric field magnitudes and local impact-generation rates varied significantly with spatial grid resolution, rendering absolute breakdown voltages mesh-dependent. Thermal modeling further revealed that while an idealized whole-body heat sink restricted lattice temperature rise to approximately four kelvins, an assumed realistic bottom-substrate thermal dissipation boundary caused internal temperatures to spike past six hundred kelvins, underscoring that physical device fabrication and thermal packaging measurements remain necessary before long-source SOI-MESFET architectures can be deployed in commercial millimeter-wave systems.
Subject of Research: TCAD numerical investigation of source-extension engineering in SOI-MESFETs for millimeter-wave amplification
Article Title: Long-source SOI-MESFET for millimeter-wave gain: A TCAD-based numerical investigation of source-extension engineering, breakdown trade-off and small-signal performance
Article References: Mohammadi, H., & Hayati, M. (2026). Long-source SOI-MESFET for millimeter-wave gain: A TCAD-based numerical investigation of source-extension engineering, breakdown trade-off and small-signal performance. Results in Engineering, 32, Article 114099. https://doi.org/10.1016/j.rineng.2026.114099
Image Credits: AI Generated
DOI: 10.1016/j.rineng.2026.114099
Keywords: SOI-MESFET, source-extension engineering, TCAD simulation, Silvaco ATLAS, millimeter-wave gain, breakdown voltage, transconductance, cut-off frequency, maximum oscillation frequency, Mason's unilateral gain, Schottky gate, series resistance, floating-body effect, impact ionization, drift-diffusion transport, parasitic capacitance, small-signal analysis, semiconductor device physics
News Source: Denise Maddox. (November 20, 2026). Simulated Source Extension Boosts SOI-MESFET High-Frequency Gain Without Sacrificing Breakdown Voltage. Scienmag.



