betaFeedback

WorksPhysics and Astronomy

A Symmetric Unified Transport and Charge Model for Metal-Oxide-Semiconductor Field-Effect Transistor from Diffusive to Ballistic Regimes

Abstract

This paper presents a symmetric unified transport (UT) compact model for metal-oxide-semiconductor field-effect transistors (MOSFETs) that bridges drift-diffusion (DD) and ballistic transport (BT) regimes. The proposed model self consistently accounts for both current and charge across the DD-BT transition. Quantum capacitance and carrier transport are incorporated into the charge density formulation. Drain side velocity saturation and the source side thermal velocity limit are unified within a single framework using a physically motivated high field scattering length, enabling accurate modeling from DD square law behavior to the ballistic limit. In addition, a physical channel charge and capacitance model is developed to capture capacitance reduction in the quasi-ballistic regime, which is not considered in standard compact models. The model is verified using theoretical analysis and experimental data from MOSFETs with multiple channel lengths, achieving accurate fitting using only physically motivated model parameters. The formulation is continuous and symmetric, and it passes both DC and AC symmetry tests.

Introduction

MOSFETs are the fundamental building blocks for modern integrated circuits (ICs). Today’s ICs may contain MOSFETs with channel lengths ranging from the micrometer scale for analog design to the nanometer scale for digital design. As the channel length decreases, carrier transport gradually transitions from drift–diffusion (DD) to ballistic transport (BT). Advanced MOSFET technologies, such as FinFET, FDSOI, and GAA, can achieve channel lengths in the sub-20-nm regime, reaching more than 80% of the nondegenerate thermal velocity with a ballistic ratio of 0.5 – 0.7 [1, 2, 3]. Therefore, it is important to develop a physical compact model that can smoothly describe transport mechanisms across a wide range of channel lengths.

Researchers have been studying the unified model including DD and BT. The virtual-source (VS) model is a widely recognized framework for describing nanotransistors based on a scattering theory [4, 5, 6, 7]. It provides an intuitive explanation of quasi-ballistic transport by introducing apparent mobility through Matthiessen’s rule, which connects ballistic and diffusive mobility. However, the basic VS model cannot accurately model long-channel MOSFETs exhibiting square-law behavior. To address this limitation, a modified VS emission–diffusion (ED) model was proposed to extend the VS formulation to long-channel devices [8]. The VSED model employs a transport-dependent critical length to describe both long-channel and short-channel devices within the VS framework.

Nevertheless, it is not compatible with the well-established DD models used in industry-standard compact models, making integration into existing frameworks difficult.

Khandelwal et al. proposed a unified compact model bridging DD and BT based on the Berkeley Short-Channel IGFET Model (BSIM) [9]. In their approach, Matthiessen’s rule is used to combine the DD velocity with the thermal velocity, enabling BT effects to be incorporated into the standard DD model. However, concerns remain regarding the physical interpretation of this formulation. The DD model is reformulated such that the current is expressed as the product of the average charge and the DD velocity, which is then combined with the thermal velocity to determine the final carrier velocity. Since the thermal velocity fundamentally limits the velocity of source charge rather than the average charge, this formulation is less physically sound.

Another approach is the distributed channel model, which divides the channel into multiple ballistic segments [10]. As the number of segments increases, the MOSFET behavior transitions from BT to DD. To reduce simulation runtime, a series connection of DD and BT MOSFET models has been proposed to reduce the number of ballistic segments. However, this approach still increases simulation time due to the introduction of additional internal nodes.

Despite these efforts, charge modeling under quasi-ballistic transport has been seldom discussed. Wei et al. demonstrated, using a simplified potential profile, that the total channel charge under BT conditions is smaller than that under DD conditions [11]. A TCAD study also showed this result [2]. However, a symmetric and unified charge model spanning DD to BT is still lacking.

In this paper, a unified transport (UT) model is proposed. Both current–voltage (IV) and charge–voltage (QV) models are developed to smoothly transition from DD to BT. The proposed model is physically sound and avoids the use of channel segmentation. It is continuous and symmetric, overcoming the limitations of conventional source-referenced quasi-ballistic models. The model is verified using MOSFET data with various channel lengths.

Unified Transport Model

The proposed MOSFET model consists of three components: A) the charge density model, B) the current model, and C) the total charge model.

Charge Density Model

A simple charge density model is chosen based on 1D Poisson and charge sheet approximation [12, 13]. The framework is generic, independent of the choice of model, and can be applied to more sophisticated charge calculations [14].

Two types of charge density models are considered in this work: the charge at the DD limit and the top-of-the-barrier charge model. Under DD conditions, the channel is near equilibrium, allowing carriers to relax to a local quasi-equilibrium state. Using the gradual channel approximation (GCA), the charge density is expressed as (1), where VgiV_{gi} is the intrinsic gate voltage, VcV_c is the channel quasi-Fermi potential, and qiq_i is the mobile charge. The threshold voltage VTV_T is given by (2), accounting for drain-induced barrier lowering (DIBL), where VdsxV_{dsx} is the effective drain–source voltage and η\eta is the DIBL factor. η\eta is treated as a constant here for simplicity but it should be length dependent due to short channel effects (SCE). VdsxV_{dsx} is a smoothed absolute VdsiV_{dsi} as (3) to ensure symmetry where VdsiV_{dsi} is the intrinsic drain-source voltage and δ\delta is the smoothing factor. CinsC_{ins} denotes the gate insulator capacitance, and CeffC_{eff} is the effective gate capacitance including quantum correction, which is assumed to follow (4) with CQC_Q as the quantum capacitance. The thermal voltage is denoted by ϕt\phi_t, and the subthreshold slope (SS) factor nn is computed by (5) to account for short-channel effects, where n0n_0 is the nominal SS factor and ndn_d is the voltage-dependent coefficient. δ\delta is set to ϕt\phi_t in this paper.

Vgi−Vc−VT=qiCeff+nϕtln⁡(qinCinsϕt)(1)V_{gi}-V_c-V_T=\frac{q_i}{C_{eff}}+n\phi_t\ln\left(\frac{q_i}{nC_{ins}\phi_t}\right) \tag*{(1)}
VT=VT0−ηVdsx(2)V_T=V_{T0}-\eta V_{dsx} \tag*{(2)}
Vdsx=Vdsi2+δ2−δ(3)V_{dsx}=\sqrt{V_{dsi}^{2}+\delta^{2}}-\delta \tag*{(3)}
Ceff=(Cins−1+CQ−1)−1(4)C_{eff}=\left(C_{ins}^{-1}+C_Q^{-1}\right)^{-1} \tag*{(4)}
n=n0+ndVdsx(5)n=n_0+n_dV_{dsx} \tag*{(5)}

The above equations assume GCA and are not applicable under BT [11]. In the ballistic regime, the channel is highly nonequilibrium, and local quasi-equilibrium cannot be assumed. Instead of using (1) to compute the mobile charge, the VS or top-of-the-barrier charge (qtopq_{top}) can be calculated using the Landauer approach [6, 7]. By combining the Landauer approach and Poisson’s equation, qtopq_{top} can be solved using (2). VtopV_{top} is the channel Fermi potential at the virtual source (VS), computed from (7), with VdiV_{di} and VsiV_{si} denoting the intrinsic drain and source voltages. This equation evaluates the minimum of VdiV_{di} and VsiV_{si}. Eq. (2a) accounts for carrier injection from both the source and drain with a transmission rate given by T=λ/(λ+L)T=\lambda/(\lambda+L), where λ\lambda is the scattering length and LL is the channel length. More rigorously, LL should be replaced by the voltage-dependent critical length ℓ\ell [5, 6, 8]. For simplicity, LL is used in this work, and the results show that this assumption does not affect the outcome because TT is not directly used to define the injection velocity of this model as the other models do [6, 8].

Vgi−Vtop−VT=qtopCeff+nϕtln⁡(2qtopnCinsϕt(2−T+Texp⁡(−Vdsx/ϕt)))(6)V_{gi}-V_{top}-V_T=\frac{q_{top}}{C_{eff}}+n\phi_t\ln\left(\frac{2q_{top}}{nC_{ins}\phi_t(2-T+T\exp(-V_{dsx}/\phi_t))}\right) \tag*{(6)}
Vtop=0.5(Vdi+Vsi−Vdsx)(7)V_{top}=0.5(V_{di}+V_{si}-V_{dsx}) \tag*{(7)}

Current Model

Similar to the core model, the current model is also divided into two parts: the DD velocity (vddv_{dd}) and the BT velocity (vbtv_{bt}).

Then, based on the Lundstrom Antoniadis model [5], under the form of transmission rate of λ/(λ+ℓ)\lambda/(\lambda+\ell), the effective injection velocity (veffv_{eff}) can be written as an inverse combination of vddv_{dd} and vbtv_{bt} of Matthiesen’s rule.

To obtain the DD current density (iddi_{dd}) is first calculated. DD-based compact models have been widely studied and share a common form, as expressed in (3) [9, 12, 15, 16]. Here, qsq_s and qdq_d are the mobile charge densities at the source and drain end. μ0\mu_0 is the low-field mobility defined as vth0λ2ϕt\frac{v_{th0}\lambda}{2\phi_t} with vth0v_{th0} being the nondegenerate thermal velocity, and UAU_A and EUE_U are the mobility-degradation parameters. β1\beta_1 is the fitting parameter for velocity saturation, and vsatv_{sat} is the saturation velocity. Unlike many models that treat vsatv_{sat} as a constant, vsatv_{sat} in this work exhibits a length dependence following (11). From the Monte Carlo simulation [17], it is observed that the drain velocity increases over the long channel limit as the channel length decreases, which is known as velocity overshoot. To model this phenomenon, a semi-physical model is proposed in [2]. It introduces an effective scattering length for saturation velocity (λsat\lambda_{sat}) to model the reflection rate of the high-field scattering (r=Lλsat+Lr=\frac{L}{\lambda_{sat}+L}). Following the physical insight that drain velocity is not bound when no scattering exists and saturates to vsat0v_{sat0} at long channel, vsatv_{sat} is modeled by vsat0/rv_{sat0}/r as (11) which resembles the empirical fitting function shown in [17]. Although not explicitly stated, the ∣qs−qd∣|q_s-q_d| in (8) is smoothed to ensure C∞C^\infty continuity.

idd=μeffL(qs2−qd2)/(2Ceff)+nϕt(qs−qd)(1+(∣qs−qd∣/(CeffVdsat1))β1)1/β1(8)i_{dd}=\frac{\mu_{eff}}{L}\frac{(q_s^2-q_d^2)/(2C_{eff})+n\phi_t(q_s-q_d)}{\left(1+\left(|q_s-q_d|/(C_{eff}V_{dsat1})\right)^{\beta_1}\right)^{1/\beta_1}} \tag*{(8)}
μeff=μ01+UA(qtop/(Ceffϕt))EU(9)\mu_{eff}=\frac{\mu_0}{1+U_A(q_{top}/(C_{eff}\phi_t))^{E_U}} \tag*{(9)}
Vdsat1=vsatL/μeff(10)V_{dsat1}=v_{sat}L/\mu_{eff} \tag*{(10)}
vsat=vsat0(1+λsatL)(11)v_{sat}=v_{sat0}\left(1+\frac{\lambda_{sat}}{L}\right) \tag*{(11)}

qsq_s and qdq_d are obtained by solving (1), which requires knowledge of VcV_c. This model adopts a widely accepted approach to determine VcV_c [15]. First, VcV_c at the electrical source is assumed to be unaffected by transport and is equal to VsiV_{si} or VdiV_{di} depending on the voltage polarity. At the electrical drain, the lower bound of VcV_c is determined using the saturation velocity vsatv_{sat} and current continuity. After accounting for MOSFET symmetry, the channel quasi-Fermi potentials at the source and drain ends can be expressed as (4). qtopxq_{topx} denotes qtopq_{top} at the DD limit. To avoid recomputing (1), it is approximated by dividing qtopq_{top} by the transmission factor. Vdsat2V_{dsat2} represents the maximum voltage across the channel, and (15) and (16) are used to smoothly clamp Vcd,sV_c^{d,s} to Vdsat2V_{dsat2} while accounting for voltage polarity, where the superscript denotes the channel position and β2\beta_2 is a fitting parameter. Vcd,sV_c^{d,s} is fed into (1) to calculate qsq_s and qdq_d.

qtopx=2qtop2−T+Texp⁡(−Vdsx/ϕt)(12)q_{topx}=\frac{2q_{top}}{2-T+T\exp(-V_{dsx}/\phi_t)} \tag*{(12)}
qmin=qtopx−CeffVdsat1(2qtopxCeffVdsat1−1)(13)q_{min}=q_{topx}-C_{eff}V_{dsat1}\left(\sqrt{\frac{2q_{topx}}{C_{eff}V_{dsat1}}-1}\right) \tag*{(13)}
Vdsat2=qtopx−qminCeff+nϕtln⁡(qtopxqmin)(14)V_{dsat2}=\frac{q_{topx}-q_{min}}{C_{eff}}+n\phi_t\ln\left(\frac{q_{topx}}{q_{min}}\right) \tag*{(14)}
Vcd,s=Vdsat2+e±Vdsi/ϕt(15)V_{c}^{d,s}=V_{dsat2}+\mathrm{e}^{\pm V_{dsi}/\phi_t} \tag*{(15)}
Vcd,s=Vsi,di±Vdsi(1+((Vdsx+δ)/Vdsat)β2)1/β2(16)V_{c}^{d,s}=V_{si,di}\pm\frac{V_{dsi}}{\left(1+\left((V_{dsx}+\delta)/V_{dsat}\right)^{\beta_2}\right)^{1/\beta_2}} \tag*{(16)}

By dividing iddi_{dd} by qtopq_{\mathrm{top}}, the DD velocity is obtained as vdd=idd/qtopv_{dd}=i_{dd}/q_{\mathrm{top}}. It is worth noting that vddv_{dd} represents the carrier velocity at the virtual source, rather than the velocity evaluated at the average channel charge qavg=(qs+qd)/2q_{\mathrm{avg}}=(q_s+q_d)/2 as assumed in [9, 18]. The injection velocity is defined at the top of the barrier. Therefore, the physically correct approach is to apply Matthiessen’s rule to combine vbtv_{bt} and vddv_{dd}, both defined at the VS, to obtain veffv_{\mathrm{eff}}. Another issue with using qavgq_{\mathrm{avg}} in the final current expression is that qavgq_{\mathrm{avg}} is not well defined under BT. Although qsq_s can still be evaluated under ballistic conditions as qtopq_{\mathrm{top}}, qdq_d can no longer be obtained from (1) due to the breakdown of the GCA [11].

The expression for vbtv_{bt} is derived next. Since quasi-ballistic effects are captured by Matthiessen’s rule, only the fully ballistic case is considered. The ballistic current density (ibti_{bt}) is expressed in (5) [7], where vthv_{th} denotes the thermal injection velocity accounting for carrier degeneracy as modeled in (5b), with α\alpha and mm as fitting parameters.

ibt=qtopvthtanh⁡(Vdsi2ϕt)(17)i_{bt}=q_{\mathrm{top}}v_{th}\tanh\left(\frac{V_{dsi}}{2\phi_t}\right) \tag*{(17)}
vth=vth0(1+α(qtopCeffϕt)m)(18)v_{th}=v_{th0}\left(1+\alpha\left(\frac{q_{\mathrm{top}}}{C_{\mathrm{eff}}\phi_t}\right)^m\right) \tag*{(18)}

Equation (5a) can be approximated and rewritten as (6a) using tanh⁡(x)≈x/(1+xβ3)1/β3\tanh(x)\approx x/(1+x^{\beta_3})^{1/\beta_3}, which closely resembles (3a). β3\beta_3 is a fitting parameter and is set to 4 in this paper, closely matching tanh. In this form, by defining vbtv_{bt} as (6b), (6a) behaves as a saturation function that clamps vbtv_{bt} at vthv_{th}.

ibt=qtopvthVdsi(1+(vth(Vdsix+δ)2ϕtvth)β3)1/β3(19)i_{bt}=q_{\mathrm{top}}\frac{v_{th}V_{dsi}}{\left(1+\left(\frac{v_{th}(V_{dsi}x+\delta)}{2\phi_t v_{th}}\right)^{\beta_3}\right)^{1/\beta_3}} \tag*{(19)}
vbt=vthVdsi2ϕt(20)v_{bt}=\frac{v_{th}V_{dsi}}{2\phi_t} \tag*{(20)}

Finally, the drain-source current IdsI_{ds} is calculated using (7) by replacing vbtv_{bt} in (6) with veffv_{\mathrm{eff}}, where WW is the channel width. veffv_{\mathrm{eff}} is obtained using Matthiessen’s rule as in (7b), and the thermal injection limit is enforced through the smoothed clamping functions in (7a). The ∣veff∣\lvert v_{\mathrm{eff}}\rvert in (7a) is also smoothed to ensure C∞C^\infty continuity. Using this formulation, the critical length ll need not be determined without a detailed band-structure description, as the carrier velocity smoothly transitions between the ballistic and DD limits via Matthiessen’s rule. In addition, the model incorporates both drain-side velocity saturation and the source-side thermal injection limit. One may notice that (7b) encounters a 0-divided-by-0 issue at Vdsi=0V_{dsi}=0. From both physics and mathematics (L'Hôpital's Rule), we can know that veff=0v_{\mathrm{eff}}=0 when Vdsi=0V_{dsi}=0 and the limits exist from both Vdsi=0+V_{dsi}=0^+ and Vdsi=0−V_{dsi}=0^-. Therefore, we can simply set the veff=0v_{\mathrm{eff}}=0 at Vdsi=0V_{dsi}=0 without worrying about the continuity which is a common practice in compact modeling.

Ids=Wqtopveff(1+(∣veff∣vth)β3)1/β3(21)I_{ds}=Wq_{\mathrm{top}}\frac{v_{\mathrm{eff}}}{\left(1+\left(\frac{\lvert v_{\mathrm{eff}}\rvert}{v_{th}}\right)^{\beta_3}\right)^{1/\beta_3}} \tag*{(21)}
veff=vbtvddvbt+vdd(22)v_{\mathrm{eff}}=\frac{v_{bt}v_{dd}}{v_{bt}+v_{dd}} \tag*{(22)}

Total Charge Model

A compact model must capture not only IV but also QV/CV characteristics. Intrinsic gate, drain, and source charges are obtained by integrating the channel charge. Under the GCA and DD conditions, the channel charge distribution is well defined and can be integrated to yield total charges, as shown in [19]. However, this approach is not applicable in the ballistic regime, where the GCA breaks down and local equilibrium cannot be assumed. In this case, the charge distribution can instead be determined by using current continuity and energy conservation while assuming a channel potential profile [11].

To connect the two limits, a physical charge model is developed as follows. qtopq_{\mathrm{top}} can be expressed as (8a), where q0q_0 is the nominal charge and Δϕ\Delta\phi is the potential deviation from the nominal value. Eq. (8a) can be reformulated as (8b) and separated into two terms. The first term represents the charge under local equilibrium, corresponding to the DD condition (qddq_{dd}), while the second term represents the charge in the ballistic limit (qbtq_{bt}). Here, the channel charge is expressed as a weighted sum of qddq_{dd} and qbtq_{bt}. This result is applied to other positions along the channel without further proof. It should be noted that this approach differs from that in [20], which employs separate carrier transmission channels for the DD and BT components. In the proposed model, the current is calculated as a lumped quantity without separating the charges. The use of (8c) does not identify which portion of the charge experiences scattering and which does not. Instead, (8c) should be interpreted as indicating that some of the carriers can relax to a local equilibrium state due to scattering. This model uses the concept of effective transmission rate, comparable to series resistance as defined by scattering theory, rather than employing separate transport channels for DD and BT as parallel resistors.

qtop=q0eΔϕϕt2−T+Te−Vdsi/ϕt2(23)q_{\mathrm{top}}=q_0e^{\frac{\Delta\phi}{\phi_t}}\frac{2-T+Te^{-V_{dsi}/\phi_t}}{2} \tag*{(23)}
q0eΔϕϕt2−2T+T+Te−Vdsi/ϕt2=q0eΔϕϕt[(1−T)+T(1+e−Vdsi/ϕt)2](24)q_0e^{\frac{\Delta\phi}{\phi_t}}\frac{2-2T+T+Te^{-V_{dsi}/\phi_t}}{2}=q_0e^{\frac{\Delta\phi}{\phi_t}}\left[(1-T)+\frac{T(1+e^{-V_{dsi}/\phi_t})}{2}\right] \tag*{(24)}
qtop=(1−T)qdd+Tqbt(25)q_{\mathrm{top}}=(1-T)q_{dd}+Tq_{bt} \tag*{(25)}

Therefore, the source- and drain-side charges under UT, qs,dutq_{s,d}^{ut}, are expressed in (9a), where qs,dq_{s,d} denotes the local-equilibrium charge obtained from (1) and (4). The superscript indicates the unified transport (ut) or the ballistic transport (bt). The subscript means the position in the channel. The ballistic charge, qs,dbtq_{s,d}^{bt}, is estimated using the method in [11]. By energy conservation, we can obtain the velocity of carrier at each channel position as (9b) where vx0v_{x0} is the velocity at the virtual source, V(x)V(x) is the potential drop in the channel, qq is the elementary charge, and m∗m^* is the effective mass. Using (9b), current continuity and assuming linear potential, the ballistic mobile charge in the channel qbt(x)q^{bt}(x) can be expressed as (9c). The parameter γ\gamma is related to vx0v_{x0} and m∗m^*, and is treated here as a fitting parameter. Based on (9c), qs,dutq_{s,d}^{ut} can be modeled by (9d & e) where Vdsi+V_{dsi}^{+} is the one-sided drain-source voltage that smoothly clamps VdsiV_{dsi} to its positive or negative component by (9f). Using that, qsbt(qdbt)q_s^{bt}(q_d^{bt}) becomes the qtop/1+γVdsiq_{\mathrm{top}}/\sqrt{1+\gamma V_{dsi}} when Vdsi<0V_{dsi}<0 (Vdsi>0V_{dsi}>0). Because carrier acceleration is not significant until VdsatV_{dsat}, Vdsi+V_{dsi}^{+} is smoothed using VdsatV_{dsat} to mimic this effect as (9g), where VdsatV_{dsat} is defined as a weighted sum of the DD Vdsat2V_{dsat2} and 2ϕt2\phi_t. 1+0.5γVdsat\sqrt{1+0.5\gamma V_{dsat}} is added to (9d & e) to make sure qs,dbt=qtopq_{s,d}^{bt}=q_{\mathrm{top}} at Vdsi=0V_{dsi}=0.

qs,dut=(1−T)qs,d+Tqs,dbt(26)q_{s,d}^{ut}=(1-T)q_{s,d}+Tq_{s,d}^{bt} \tag*{(26)}
vx=vx02+2qV(x)m∗(27)v_x=\sqrt{v_{x0}^{2}+\frac{2qV(x)}{m^{*}}} \tag*{(27)}
qibt(x)=qibt(0)1+γVdsix/L(28)q_i^{bt}(x)=\frac{q_i^{bt}(0)}{\sqrt{1+\gamma V_{dsi}x/L}} \tag*{(28)}
qsbt=qtop1+0.5γVdsat1−γVdsi(29)q_s^{bt}=\frac{q_{\mathrm{top}}\sqrt{1+0.5\gamma V_{dsat}}}{\sqrt{1-\gamma V_{dsi}}} \tag*{(29)}
qdbt=qtop1+0.5γVdsat1+γVdsi(30)q_d^{bt}=\frac{q_{\mathrm{top}}\sqrt{1+0.5\gamma V_{dsat}}}{\sqrt{1+\gamma V_{dsi}}} \tag*{(30)}
Vdsi∓=0.5(Vdsi∓Vdsi2+Vdsat2)(31)V_{dsi}^{\mp}=0.5\left(V_{dsi}\mp\sqrt{V_{dsi}^{2}+V_{dsat}^{2}}\right) \tag*{(31)}
Vdsat=(1−T)Vdsat2+2Tϕt(32)V_{dsat}=(1-T)V_{dsat2}+2T\phi_t \tag*{(32)}

The total intrinsic gate and drain charges, (QGi,QDi)(Q_{Gi}, Q_{Di}), are obtained by substituting quts,dq_{u t}^{s,d} into the DD charge model in [19]. However, since that model is derived under GCA, corrections are required for ballistic and quasi-ballistic conditions. It is found that the charge distribution and total charge can be reduced by the factors given in (9). With this modification, the conventional DD charge model can be extended to a UT charge model that accounts for BT. The final expression is shown as (10) where qia=0.5(quts+qutd)q_{ia}=0.5(q_{ut}^{s}+q_{ut}^{d}), dq=quts−qutddq=q_{ut}^{s}-q_{ut}^{d}, Kq=2(qia+Ceffnϕt)K_q=2(q_{ia}+C_{eff}n\phi_t), and Kbt=1+TbaγVdsxK_{bt}=\sqrt{1+T^{b}a\gamma V_{dsx}}.

QGi=WLKbt(qia+dq26Kq)(33)Q_{Gi}=\frac{WL}{K_{bt}}\left(q_{ia}+\frac{dq^{2}}{6K_q}\right) \tag*{(33)}
QDi=−WLKbt(qia−dq6(1−dqKq(1+dq5Kq)))(34)Q_{Di}=-\frac{WL}{K_{bt}}\left(q_{ia}-\frac{dq}{6}\left(1-\frac{dq}{K_q}\left(1+\frac{dq}{5K_q}\right)\right)\right) \tag*{(34)}
QSi=−QGi−QDi(35)Q_{Si}=-Q_{Gi}-Q_{Di} \tag*{(35)}

KbtK_{bt} is a factor introduced to account for BT effects, where VdsxV_{dsx} follows the same formulation as (1c), with VdsatV_{dsat} used as the smoothing factor. KbtK_{bt} is based on 1+γVdsi\sqrt{1+\gamma V_{dsi}} from (9d & e), although the actual factor is position dependent as (9c). Thus, to integrate it with the DD model with minimal change, a fitting parameter aa is added to compensate for discrepancy by assuming that the charge distribution is uniformly scaled by KbtK_{bt}. To consider the length dependency and ballistic ratio, TbT^{b} is further added with TT as the transmission factor and bb is a fitting parameter. Both aa and bb are empirical parameters to incorporate BT effects in the DD charge model.

Validation

First, the model is validated against theoretical predictions by comparison with ideal DD and BT models. Figure 1 to 3 shows the model simulations compared to the ideal drift-diffusion model and the ballistic model with no 2D electrostatics, SCE, including DIBL, VTV_T roll-off, and SS degradation to isolate transport effects. For this study, λ=16\lambda=16 nm, λsat=10\lambda_{sat}=10 nm, γ=30\gamma=30, a=0.28a=0.28, and b=1b=1 are assumed.

Figure 1. (a) IdVd and (b) CggiVg curves of the DD and UT model at L=2L=2 μm.

Figure 1 verifies the model behavior in the long-channel limit by comparing the IdVd and CggiVg (intrinsic gate capacitance) curves with those of the DD model. At L=2L=2 μ\mum, and the device operates in the fully diffusive regime. In this case, (7a) reduces to vddv_{dd}, and the channel charge fully relaxes to local equilibrium. Consequently, the UT model reproduces the classic square-law behavior, yielding the same IV and CV curves as the DD model. When the channel length is reduced to 30 nm, λ\lambda becomes comparable to LL, and the MOSFET enters the quasi-ballistic regime, where carriers can travel beyond the long-channel saturation velocity. The simulated IV and CV characteristics from the DD, BT, and UT models are shown in Figure 2. The DD model employs an apparent mobility but does not include the length-dependent vsatv_{sat}, allowing the velocity overshoot effect captured by the UT model to be highlighted.

Figure 2. (a) IdVd and (b) CggiVg curves of the DD, BT and UT model at L=30L=30 nm.

As shown in Figure 2a, the current predicted by the DD model is limited by vsat0v_{sat0}, whereas the UT model exceeds this limit and approaches the ballistic regime due to the scattering-aware vsatv_{sat} formulation in (3d). Figure 2b presents the corresponding CV results. In the DD model, the ratio between the high-VdV_d and low-VdV_d CV curves decrease with channel length as a result of velocity saturation. This behavior is no longer valid in the quasi-ballistic regime, where carriers can exceed the saturation velocity near the drain, leading to channel charge reduction [2]. With (9) and (10), the UT model produces CV characteristics that lie between the DD and BT limits.

At L=1L=1 nm, the channel exhibits more than 95% ballisticity. The effective velocity veffv_{\mathrm{eff}} of the UT model is close to vbtv_{bt}, indicating that the device operation is limited by source injection rather than drain-side velocity saturation. Figure 3 shows the characteristic curves under this condition. Both the IV and CV characteristics predicted by the UT model closely approach the ballistic limit.

Figure 3. (a) IdVd and (b) CggiVg curves of the BT and UT model at L=1L=1 nm.

After the theoretical validation, the model is also verified with experimental data. Figure 4 to 7 demonstrates the fitting results of ETSOI fabricated by the same IBM process technology from L=980L=980 nm to 30 nm [8, 21]. For all IdVd curves, the VgsV_{gs} are biased from -0.2 to 0.5 V with 0.1 V-step. For all IdVg curves, VdsV_{ds} are biased at 0.01 V, 0.05 V and 1 V. A single set of universal transport parameters with constant vthv_{th} is summarized in Table I, using a source/drain resistance of 125 Ωμ\Omega\mum. The injection velocity at 30 nm is calculated to be 8.6×1068.6 \times10^{6} cm/s with a ballistic ratio of 0.75 that closely matches the values reported in [1, 3]. This shows that the proposed model can correctly predict the transport from diffusive to ballistic regime using a single set of physically meaningful transport parameters. Although this work focuses on length dependent transport physics and modeling of MOSFETs, electrostatic, DIBL and SCE parameters also have length dependency due to the 2D electrostatic effects. VT0V_{T0}, η\eta, n0n_0, and ndn_d are subjected to the effect of VTV_T roll-off, DIBL, SCE, and SS degradation. On the other hand, β2\beta_2 representing how the drain potential is saturated is also affected by the 2D electrostatics and is length dependent. The author follows the presentation in [8] to treat those electrostatic parameters as individual constants for different channel length, which are summarized in Table II. The length dependent models for those electrostatic parameters are widely studied by compact models like BSIM [22]. Those existing models can be applied to the parameters in this paper with minimal modification.

Figure 4. (a) IdVd and (b) IdVg of a 980 nm ETSOI. Symbol: data [8]. Line: model.

λ\lambda (nm)λsat\lambda_{sat} (nm)vsat0v_{sat0} (cm/s)vthv_{th} (cm/s)UA
15.898.5E61.14E74.4E-5
EUβ1\beta_1β3\beta_3CinsC_{ins} (F/m²)CQC_Q (F/m²)
2.5440.0310.06

Table I. Universal fitting parameters for ETSOI

30 nm40 nm50 nm980 nm
VT0V_{T0} (mV)-150-150-143-120
η\eta (mV/V)5545305
n0n_01.51.351.31.12
ndn_d (1/V)0.250.20.10
β2\beta_211.528

Table II. Individual fitting parameters for ETSOI

IdVd and IdVg of a 50 nm ETSOI

Figure 5. (a) IdVd and (b) IdVg of a 50 nm ETSOI. Symbol: data [21]. Line: model.

IdVd and IdVg of a 40 nm ETSOI

Figure 6. (a) IdVd and (b) IdVg of a 40 nm ETSOI. Symbol: data [21]. Line: model.

IdVd and IdVg of a 30 nm ETSOI

Figure 7. (a) IdVd and (b) IdVg of a 30 nm ETSOI. Symbol: data [21]. Line: model.

Furthermore, the model is verified with an advanced FinFET [23], as shown in Figure 8. Since the gate length and EOT are not reported in [23], a 16 nm channel length is chosen based on the technology node. By fitting the low VDSV_{\mathrm{DS}} curve with Ceff=0.043C_{\mathrm{eff}}=0.043 F/m2^2, the inversion charge at VGS=0.7V_{\mathrm{GS}}=0.7 V is approximately 1.7 μC/cm2^2. To achieve an on-current of 2300 μA/μm, an injection velocity of 1.35×1071.35 \times10^7 cm/s is required, which exceeds the nondegenerate thermal velocity of silicon. Therefore, the degenerate vthv_{\mathrm{th}} model is used to capture this data, resulting in a maximum vthv_{\mathrm{th}} of 1.9×1071.9 \times10^7 cm/s and a veffv_{\mathrm{eff}} of 1.44×1071.44 \times10^7 cm/s from the model. The fitting parameters are summarized in Table III.

λ\lambda (nm)λsat\lambda_{\mathrm{sat}} (nm)vsat0v_{\mathrm{sat0}} (cm/s)vth0v_{\mathrm{th0}} (cm/s)
10101.8E71.14E7
α\alphamUAEU
0.30.37E-42.5
CeffC_{\mathrm{eff}} (F/m2^{2})VT0V_{\mathrm{T0}} (mV)η\eta (mV/V)n0n_0
0.043270221.12
ndn_d (1/V)β1\beta_1β2\beta_2β3\beta_3
0.032.544

Table III. Fitting parameters for FinFET

Figure 8. (a) IdVdI_dV_d and (b) IdVgI_dV_g of an Intel 3 FinFET. Symbol: data [20]. Line: model. (a) VgsV_{gs} are biased from 0.2 to 0.7 V with 0.1 V-step. (b) VdsV_{ds} are biased at 0.05 V and 0.7 V.

The model is developed with special attention to device symmetry. DC and AC symmetry tests are standard evaluations of MOSFET symmetry, smoothness, and continuity around Vds=0V_{\mathrm{ds}}=0 [24]. Most source-referenced models suffer from discontinuity at Vds=0V_{\mathrm{ds}}=0. Although the formulation of velocity saturation and ballistic transport inherently relies on source referencing due to the use of virtual-source charge for current and charge calculations, the proposed model avoids discontinuity and nonsmoothed behavior by introducing appropriate smoothing functions. As a result, the model successfully passes both DC and AC symmetry tests, as shown in Figure 9, which is typically not satisfied by source-referenced quasi-ballistic transport models.

Figure 9. The DC and AC symmetry test of this model.

Conclusion

A unified transport model for FETs is developed to describe carrier transport from the diffusive to the ballistic regime. The model improves conventional approaches for combining DD and BT velocities by introducing a physically consistent treatment of the thermal injection limit and velocity overshoot. A quasi-ballistic charge model is also derived to account for the transition between nonequilibrium and quasi-equilibrium charge distributions. The developed current and charge models are compatible with existing DD-based compact models with simple modifications. The model is verified using both theoretical and experimental data, demonstrating its predictive capability.

Acknowledgement

The author would like to thank Prof. Chenming Hu for his mentorship, and Prof. Dimitri Antoniadis for his constructive and valuable feedback on this work. This work was conducted independently and does not represent the views or intellectual property of Synopsys.

References

  1. [1]Y. -H. Liao, K. A. Aabrar, W. Chakraborty, W. Li, S. Datta, and S. Salahuddin, “Large Injection Velocities in Highly Scaled, Fully Depleted Silicon on Insulator Transistors,” IEEE Electron Device Lett., vol. 43, no. 2, pp. 184–187, Feb. 2022, doi: 10.1109/LED.2021.3153407.DOI
  2. [2]C.-T. Tung, “Study and design nanotransistor using SEMIDV device simulator,” APL Electron. Devices, vol. 1, no. 3, p. 036114, Jul. 2025, doi: 10.1063/5.0280961.DOI
  3. [3]A. Majumdar and D. A. Antoniadis, “Analysis of Carrier Transport in Short-Channel MOSFETs,” IEEE Trans. Electron Devices, vol. 61, no. 2, pp. 351–358, Feb. 2014, doi: 10.1109/TED.2013.2294380.DOI
  4. [4]A. Khakifirooz, O. M. Nayfeh, and D. Antoniadis, “A Simple Semiempirical Short-Channel MOSFET Current–Voltage Model Continuous Across All Regions of Operation and Employing Only Physical Parameters,” IEEE Trans. Electron Devices, vol. 56, no. 8, pp. 1674–1680, Aug. 2009, doi: 10.1109/TED.2009.2024022.DOI
  5. [5]M. S. Lundstrom and D. A. Antoniadis, “Compact Models and the Physics of Nanoscale FETs,” IEEE Trans. Electron Devices, vol. 61, no. 2, pp. 225–233, 2014, doi: 10.1109/TED.2013.2283253.DOI
  6. [6]S. Rakheja, M. S. Lundstrom, and D. A. Antoniadis, “An Improved Virtual-Source-Based Transport Model for Quasi-Ballistic Transistors—Part I: Capturing Effects of Carrier Degeneracy, Drain-Bias Dependence of Gate Capacitance, and Nonlinear Channel-Access Resistance,” IEEE Trans. Electron Devices, vol. 62, no. 9, pp. 2786–2793, Sep. 2015, doi: 10.1109/TED.2015.2457781.DOI
  7. [7]A. Rahman and M. S. Lundstrom, “A compact scattering model for the nanoscale double-gate MOSFET,” IEEE Trans. Electron Devices, vol. 49, no. 3, pp. 481–489, Mar. 2002, doi: 10.1109/16.987120.
  8. [8]S. Rakheja, M. Lundstrom, and D. Antoniadis, “A physics-based compact model for FETs from diffusive to ballistic carrier transport regimes,” in 2014 IEEE International Electron Devices Meeting, Dec. 2014, pp. 35.1.1–35.1.4. doi: 10.1109/IEDM.2014.7047172.DOI
  9. [9]S. Khandelwal et al., “Unified Compact Model Covering Drift-Diffusion to Ballistic Carrier Transport,” IEEE Electron Device Lett., vol. 37, no. 2, pp. 134–137, Feb. 2016, doi: 10.1109/LED.2015.2507519.DOI
  10. [10]G. Mugnaini and G. Iannaccone, “Physics-based compact model of nanoscale MOSFETs-Part I: transition from drift-diffusion to ballistic transport,” IEEE Trans. Electron Devices, vol. 52, no. 8, pp. 1795–1801, Aug. 2005, doi: 10.1109/TED.2005.851827.DOI
  11. [11]L. Wei, O. Mysore, and D. Antoniadis, “Virtual-Source-Based Self-Consistent Current and Charge FET Models: From Ballistic to Drift-Diffusion Velocity-Saturation Operation,” IEEE Trans. Electron Devices, vol. 59, no. 5, pp. 1263–1271, May 2012, doi: 10.1109/TED.2012.2186968.DOI
  12. [12]C. C. Enz and E. A. Vittoz, Charge-based MOS transistor modeling: the EKV model for low-power and RF IC design. John Wiley & Sons, 2006.
  13. [13]H. K. Gummel and K. Singhal, “Inversion charge modeling,” IEEE Trans. Electron Devices, vol. 48, no. 8, pp. 1585–1593, Aug. 2001, doi: 10.1109/16.936564.DOI
  14. [14]J. P. Duarte et al., “BSIM-CMG: Standard FinFET compact model for advanced circuit design,” presented at the ESSCIRC Conference 2015 - 41st European Solid-State Circuits Conference (ESSCIRC), Sep. 2015, pp. 196–201, doi: 10.1109/ESSCIRC.2015.7313862.DOI
  15. [15]K. Joardar, K. K. Gullapalli, C. C. McAndrew, M. E. Burnham, and A. Wild, “An improved MOSFET model for circuit simulation,” IEEE Trans. Electron Devices, vol. 45, no. 1, pp. 134–148, Jan. 1998, doi: 10.1109/16.658823.DOI
  16. [16]U. Radhakrishna, P. Choi, and D. A. Antoniadis, “Facilitation of GaN-Based RF- and HV-Circuit Designs Using MVS-GaN HEMT Compact Model,” IEEE Trans. Electron Devices, vol. 66, no. 1, pp. 95–105, Jan. 2019, doi: 10.1109/TED.2018.2848721.DOI
  17. [17]R. Granzner et al., “Simulation of nanoscale MOSFETs using modified drift-diffusion and hydrodynamic models and comparison with Monte Carlo results,” Microelectron. Eng., vol. 83, no. 2, pp. 241–246, Feb. 2006, doi: https://doi.org/10.1016/j.mee.2005.08.003.
  18. [18]A. Dasgupta, A. Agarwal, and Y. S. Chauhan, “Unified Compact Model for Nanowire Transistors Including Quantum Effects and Quasi-Ballistic Transport,” IEEE Trans. Electron Devices, vol. 64, no. 4, pp. 1837–1845, Apr. 2017, doi: 10.1109/TED.2017.2672207.DOI
  19. [19]Y. S. Chauhan et al., “BSIM6: Analog and RF Compact Model for Bulk MOSFET,” IEEE Trans. Electron Devices, vol. 61, no. 2, pp. 234–244, Feb. 2014, doi: 10.1109/TED.2013.2283084.DOI
  20. [20]A. Dasgupta, A. Agarwal, S. Khandelwal, and Y. S. Chauhan, “Compact Modeling of Surface Potential, Charge, and Current in Nanoscale Transistors Under Quasi-Ballistic Regime,” IEEE Trans. Electron Devices, vol. 63, no. 11, pp. 4151–4159, Nov. 2016, doi: 10.1109/TED.2016.2630540.DOI
  21. [21]S. Rakheja, M. S. Lundstrom, and D. A. Antoniadis, “An Improved Virtual-Source-Based Transport Model for Quasi-Ballistic Transistors—Part II: Experimental Verification,” IEEE Trans. Electron Devices, vol. 62, no. 9, pp. 2794–2801, Sep. 2015, doi: 10.1109/TED.2015.2457872.DOI
  22. [22]Y. S. Chauhan et al., FinFET GAA Modeling for IC Simulation and Design (Second Edition). Academic Press, 2024.
  23. [23]W. Hafez et al., “An Intel 3 Advanced FinFET Platform Technology for High Performance Computing and SOC Product Applications,” in 2024 IEEE Symposium on VLSI Technology and Circuits (VLSI Technology and Circuits), Jun. 2024, pp. 1–2. doi: 10.1109/VLSITechnologyandCir46783.2024.10631513.DOI
  24. [24]C. C. Mcandrew, “Validation of MOSFET model Source–Drain Symmetry,” IEEE Trans. Electron Devices, vol. 53, no. 9, pp. 2202–2206, 2006, doi: 10.1109/TED.2006.881005.DOI

Paper details

vlnc:W_9ssf286r
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents