2024 年 1 月寒风吹彻
半导体器件原理 2.6 Subthreshold Region Current and Effect of Body Bias

半导体器件原理 2.6 Subthreshold Region Current and Effect of Body Bias

2025 年 11 月 24 日

About the subthreshold behaviors of MOSFETs, the turn-on characteristics, comparison between MOSFETs and BJTs, and the effect of body bias and substrate depletion charge on MOSFET I-V characteristics.


MOSFET Subthreshold Region

For a properly designed MOSFET, the VT is usually positive for N MOSFET, and negative for P MOSFET. We will assume it is the case for the rest of this section, even though there are some exceptions.

We have derive the current equation for MOSFET when VG>VT. Consider NMOSFETs. When 0<VG<VT, we usually say that the MOSFET is operating in the subthreshold region. Many people assumes the current in this region is zero, but it is not true. In fact, the subthreshold current is very important, and determines how well a MOSFET can be turned off in modern electronic circuits.

  • Consider the potential barrier Vbi between the source-body junction
  • When VG=VT, the potential barrier almost disappears, and the MOSFET turns on
  • Before the MOSFET turns on, it is similar to a BJT, with VG changing the barrier height, and the change of barrier height is determined by the surface band bending ϕs
  • In a BJT, the collector current under forward active mode is given byIC=IC0eqVBEkT
  • Similarly, the subthreshold current of a MOSFET can be expressed asID=ID0eqϕskTwhere ID0 is a constant, and is usually measured instead of calculated
  • When VG>VT, ID is limited by the resistance of the channel, instead of the source to body potential barrier
  • From previous section, we know that in depletion region,ϕS=1n(VGVFB)
  • The subthreshold region also operates in depletion region, as we have 0<VG<VT, so we can substitute ϕS into the equation of ID, and getID=ID0eq(VGVFB)nkT
  • VFB is also a bias independent constant for a given MOSFET, it can be lumped into ID0, and we can finally getID=ID0eqVGnkT

Turn-on Characteristics

We have now obtained all the equations for drain current of a MOSFET from subthreshold to strong inversion, we can now study the characteristics of a MOSFET when it is turned on or off with VG.

Plotting the IDVG characteristics of a NMOSFET in linear scale:

I_D to V_G characteristics of a NMOSFET in linear scale

  • Before VG reaches VT, the MOSFET is in subthreshold region, and ID increases exponentially with VG
    • Because the current is relatively small in this region, it is difficult to observe its characteristics in a linear scale
  • After VG exceeds VT, but before it reaches VD+VT, the MOSFET is in the saturation region, and ID has a square dependence on VG
  • After VG exceeds VD+VT, the MOSFET is in the linear region, and ID increases linearly with VG
  • When measured with a larger VD
    • The curve remains more or less the same in subthreshold and saturation region, because VD has no effect on IDsat besides insignificant contributions of channel length modulation effect
    • The linear region extends further, as the transition point VG=VD2+VT shifts right
  • There are some similarities between this graph, and the IV characteristics of BJTs and PN junctions
    • The turn-on voltage for silicon junctions is assumed to be 0.7V
  • To observe the subthreshold characteristics more clearly, we can plot the same data in a semi-log scale
  • The subthreshold region now becomes a straight line, showing the exponential dependence of ID on VG
  • It is similar to the BJT Gummel Plot, or the logICVBE graph of a BJT
    • The part beyond Von of a BJT cannot be used, as it will cause a large current flowing through the base
    • In MOSFET, we can use this region, as the gate blocks the current with the insulating oxide
  • The slope of the subthreshold region isslope=qnkTlog(e)It measures how abrupt a MOSFET can be turned on, but the swing is more commonly usedSubthreshold Swing=1slope=nkTqln(10)(mV/dec)
    • At room temperature, the subthreshold swing is approximately 60nmV/dec, where n is the ideality factor given byn=1+CDCox

Subthreshold Swing

The subthreshold swing indicates the ratio between the on state current and off state current of a MOSFET. This is because once VT is fixed, the leakage current at VG=0 is determined by how fast VG can turn off the current, controlled by the subthreshold swing.

  • For example, S=80mV/dec, and VT=0.8V
    • This means for every 80mV decrease in VG, ID decreases by a factor of 10
    • From VG=0.8V to VG=0V, there are 10 such steps, so the current decreases by a factor of 1010
    • If VT is reduced to 0.4V, the current will only decrease by a factor of 105, 105 times larger than before, which is significant
  • The actual leakage current may be larger than the predicted value due to other effects
    • The leakage current from the drain to substrate may define the lowest bound of the leakage current
      • This current is independent of VG, and appears to be flat in the logIDVG plot
  • As S=60nmV/dec, we need to minimize n to build a good switch
    • n is given byn=1+CDCox
    • This can be done by maximizing Cox and minimizing CD
      • Reducing CD by lowering the substrate doping concentration is limited by another constraint, which will be discussed later
      • It is more common to increase Cox by using a thinner oxide layer
  • The best achievable subthreshold swing at room temperature is approximately 60mV/dec, when n=1, just the case of BJTs

Similar to the linear plot, ID measured under different VD overlaps in the subthreshold and saturation region, and start to separate in the linear region.

Current at the Threshold Voltage

Combining the subthreshold current equation and the strong inversion equations, we will observe a discontinuity at VG=VT, as the strong inversion current equation predicts ID=0 at VT.

This discontinuity occurs because we used Q=CoxVGVT to calculate the inversion charges at the channel near the source end, which means the inversion charge below threshold is zero, and abruptly appears when VG exceeds VT.

The discontinuity at V_T and why it occurs

In reality, the inversion charge appears before threshold, as there are always electrons in the conduction band to prevent the current at VT from going to 0.

Handling the current at VT requires tedious mathematics to solve the surface potential, and there are some other more advanced courses that deal with the formulation of a continuous current voltage equations from the subthreshold to the strong inversion regions.

The main takeaway of this section is that there is a small region around VT that we do not know how to calculate the current.

MOSFET v.s. BJT

BJTs and MOSFETs are usually considered very different devices operating with different principles. However, are actually very similar. They are both comprised of the sam PNP / NPN structure, and MOSFETs in subthreshold region operates very similarly to BJTs.

When considering the performance of a device, we not only consider its output, but also the loading device introduced to operate it. More specifically, the speed of a device is determined by the speed to charge the input capacitance of similar devices to the required voltage through its current.

If input capacitance is C, and has to be charged to voltage V with a driving current I to achieve transition, the delay is given by

delay=CVI

Or the speed can be characterized by IC for given power supply.

  • Consider a BJT driving itself Driving a BJT with itself
    • Assume a specific current density J0 through the BJT
    • The size of a BJT is mainly determined by the emitter area given by W and L
    • The current flows vertically, and is given byI=J0WLThe current density J_0
    • The input capacitance is mainly determined by the base-emitter junction capacitance Cπ
    • The capacitance is given byC=Cπ0WLwhere Cπ0 is the normalized base-emitter capacitance per unit area
    • The speed:speedIC=J0WLCπ0WL=J0Cπ0
    • Reducing the size of the BJT does not affect its speed, as both current and capacitance scale with area
  • Consider a MOSFET driving itself Driving a MOSFET with itself
    • Assume it has the same current density J0
    • The input capacitance is CoxWL
    • However, the current flows horizontally through the channel, with a cross-sectional area of Wtinv, where tinv is the thickness of the inversion layer
    • The current is given byI=J0WtinvThe driving current in MOSFET
    • The speed:speedIC=J0WtinvCoxWL=J0tinvCoxL
    • Because tinv is in the order of a few nanometers, which is much smaller than L, it makes a MOSFET very slow to drive itself
    • To increase the speed, we have to either increase the current drive, or decrease the loading capacitance
      • This is why we operate MOSFETs at a higher VG to provide more current flow
    • Early day MOSFETs operating in subthreshold region have such a low driving current that they are considered not usable for any meaningful applications
    • When we scale down the MOSFET, L decreases, decreasing the capacitance and increasing the speed, making size reduction advantageous for MOSFETs
    • In SOTA MOSFETs, L and tinv are getting very close, enabling MOSFET circuits to be used in subthreshold region
    • The increase in speed when scaling down is mainly contributed by the reduction of loading capacitance, instead of the increase in current, making MOSFETs suitable for integrated circuits with closely packed transistors and small parasitic capacitive loading
    • When driving external elements with high capacitive load, BJTs with large cross-sectional area is still more desirable
      • This is why BJTs are more popular as a discrete element to function as a driver for large external loads

I-V Characteristics with Substrate Bias

Up to now, we have assumed that the source and substrate of a MOSFET are connected together and grounded. However, in some applications, like source follower circuits, source voltage may be higher than the substrate voltage, or effectively a negative substrate bias is applied to the MOSFET.

A source follower

  • When source and body voltages are different, we need to pick a reference
    • In the source follower circuit, we can pick the source voltage as reference, and VT is labeled as VTB with respect to the body voltage
      • VG, VD, and VS are also labeled as VGB, VDB, and VSB respectively, to indicate they are measured with respect to the body voltage
      • The CV characteristics now is C-V characteristics with body voltage taken as referenceΔVT=γ(2ϕB+VSB2ϕB)
    • Or we can pick the source voltage as reference
      • Label VT VG, VD, and VB as VTS, VGS, VDS, and VBS
      • In MOSFETs, we are more interested in the inversion electrons in the channel, and these electrons come from the source, thus this reference is more commonly used
      • The CV characteristics now is C-V characteristics with source voltage taken as reference Its effect is mainly the change in VT
      • The IV and logIV graph will be shifted right by ΔVTThe shifted I-V and logI-V graph
      • As the capacitance with body bias in the depletion mode is smaller compared to no body bias, and the capacitance is a series of Cox and CD, thus CD decreases (because the body voltage increases the depletion width at the same gate voltage). n=1+CD/Cox, thus n decreases, and the subthreshold slope becomes steeper with negative body bias, but the lower limit of ID may still be subjective to the drain junction leakage

Substrate Depletion Charge Effect

When deriving the current equations of MOSFETs, we have assumed VT to be constant along the channel.

ID=Q(y)v(y)Q(y)=CoxW(VGVTV(y))

However, this is not true, and VT is a function of the channel voltage. We may also consider the body effect:

VT(V(y))=VFB0+2ϕB+γ2ϕB+V(y)VBγ=2qNAεSiCox

Putting it back to ID, and follow the previous derivation steps, we have

0LIDdy=0VDCoxW(VG(VFB0+2ϕB+γ2ϕB+V(y)VB)V(y))μdVID=μCoxWL((VGVFB02ϕB)VDVD2223γ((2ϕBVB+VD)3/2(2ϕBVB)3/2))

Again, we assume the range of V(y) is limited, and we can approximate VT(V(y)) as

VT(V(y))VFB0+2ϕB+γ2ϕBVB+γ22ϕBVBV(y)=VTS+γV(y)22ϕBVB

and

Q(y)=CoxW(VGVTS(1+γ22ϕBVB)V(y))=CoxW(VGVTSαV(y))where α=1+γ22ϕBVB

Following the previous derivation steps, we can get

ID=μCoxWL((VGVT)VDαVD22)

Usually, 1<α<2, and is called the substrate / bulk charge factor, which determines how strong VT varies along the channel.

ID=μCoxWL((VGVT)VDαVD22)=μCoxWL(VG(VT+α12VD)VD2)VD=μCoxWL(VGVT(new)VD2)VDwhere VT(new)=VT+α12VD

which means the equation assumes VT increases linearly with VD.

After modifying the linear region, we also need to modify the saturation region, just by finding the peek of the quadratic equation.

IDsat=12αμCoxWL(VGVT)2VDsat=VGVTα

In traditional long channel transistors, α1.5. By ignoring the VT variations along the channel, we may overestimate the current by 50%

No matter the value of α, the point (VDsat,IDsat) lies on the same straight line joining the origin and the point when α=1, because VDsat and IDsat are both scaled down by α.

I_Dsat and V_Dsat scaled by same factor

The change in VT only affects the current calculations, but not the turn-on characteristics, as they are determined by the VT at the source.


文件历史

content: principle of semi devices 2.6 c5f7548
2025 年 11 月 24 日 12:58djdjz7
© 2023 - 2025
Unless otherwise stated, text contents are licensed under CC BY-NC 4.0, and codes are licensed under MIT.
Images may be subject to separate licenses, see the image captions for details if applicable.