Threshold and the regions

Read a transistor's output curves, name its three regions, extract a threshold voltage by two methods, and measure how length, drain voltage and body bias move it.

Lesson
E2.2
Time
30 min
Runs on
ngspice
Check
5 questions

A designer took a threshold of about 0.56 V from a long transistor and biased a minimum-length one of the same width-to-length ratio at 0.6 V. The long device carries 5.08 µA there; the short one carried 1.08 µA, under a quarter.

DC sweepSame W/L, two lengths: drain current against gate voltageW/L = 2/0.15 and 13.33/1, both at VDS = 0.9 V. The log axis shows the small currents near the threshold.
  • id_short
  • id_long
1 pA100 pA10 nA1 µA100 µADrain current00.2 V0.4 V0.6 V0.8 V1.0 V1.2 VGate-source voltage VGSbias: 0.6 VL = 0.15 µm: 1.08 µAL = 1 µm: 5.08 µA1 pA100 pA10 nA1 µA100 µADrain current00.5 V1.0 VGate-source voltage VGSbias: 0.6 VL = 0.15 µm: 1.08 µAL = 1 µm: 5.08 µA
  1. Gate-source voltage VGS 600 mVid_short = 1.077 µA · id_long = 5.081 µA

    At 0.6 V the long device carries 5.08 µA and the short one 1.08 µA. The curves have the same shape, shifted along the gate axis: the short device starts to conduct later.

  2. Gate-source voltage VGS 1.2 Vid_short = 320.9 µA · id_long = 582.6 µA

    At 1.2 V the gap has narrowed but not closed: 321 µA against 583 µA. E2.3 explains the rest of the difference.

Read the plot as a table
Gate-source voltage VGSid_shortid_long
0 V1.318 pA1.856 pA
200 mV64.06 pA221.5 pA
400 mV9.511 nA46.21 nA
600 mV1.077 µA5.081 µA
800 mV31.36 µA85.18 µA
1 V152.3 µA288.1 µA
1.2 V320.9 µA582.6 µA

Measured on ngspice 42, SKY130 tt (601 points).

The short device's threshold was about 0.12 V higher. A threshold belongs to one device size at one bias, measured one particular way.

Output curves and three regions

The characterization benchOne NMOS with a source on its gate and another on its drain. The drain source's current is the drain current.
The characterization bench++VDSsweptM1W/L = 2/1VGS0.9 … 1.8 Vdg
  1. Gate drive

    VGS, the gate-to-source voltage, sets how strongly the channel forms. It stays fixed for each curve.

  2. Drain sweep

    VDS is swept from 0 to 1.8 V. The drain current ID flows from the source VDS into the drain, through the channel to ground.

    ID

Read the circuit as text
  • VDS voltage source (swept): + → d, − → 0
  • M1 NMOS transistor (W/L = 2/1): gate → g, drain → d, source → 0
  • VGS voltage source (0.9 … 1.8 V): + → g, − → 0

The body is tied to the source, at ground. The figure below runs four such devices at once, one per gate voltage.

Fix the gate voltage, sweep the drain, and record the drain current: that is one output curve. The figure draws four, for VGS = 0.9, 1.2, 1.5 and 1.8 V on a W/L = 2/1 µm NMOS.

DC sweepOutput curves of an NMOS, W/L = 2/1 µmID against VDS at four gate voltages. The marked points are where the model reports VDsat.
  • id09
  • id12
  • id15
  • id18
-50 µA050 µA100 µA150 µA200 µA250 µA300 µADrain current ID00.2 V0.4 V0.6 V0.8 V1.0 V1.2 V1.4 V1.6 V1.8 VDrain-source voltage VDS0.26 V0.44 V0.60 V0.74 V-100 µA0100 µA200 µA300 µADrain current ID00.5 V1.0 V1.5 VDrain-source voltage VDS0.26 V0.44 V0.60 V0.74 V
  1. Drain-source voltage VDS 50 mVid09 = 6.69 µA · id12 = 13.81 µA · id15 = 19.87 µA · id18 = 24.61 µA

    At small VDS every curve is a straight line through zero: the channel acts as a resistor whose value the gate sets. This is the triode region.

  2. Drain-source voltage VDS 259.6 mVid09 = 20.9 µA · id12 = 56.07 µA · id15 = 87.6 µA · id18 = 113 µA

    At 0.26 V the VGS = 0.9 V curve reaches its saturation voltage VDsat. Beyond it the current grows only slowly with the drain voltage.

  3. Drain-source voltage VDS 740.3 mVid09 = 23.69 µA · id12 = 80.74 µA · id15 = 155.9 µA · id18 = 231.8 µA

    The VGS = 1.8 V curve needs 0.74 V before it saturates, so a larger gate drive moves the knee to the right.

  4. Drain-source voltage VDS 1.7 Vid09 = 25.04 µA · id12 = 85.56 µA · id15 = 168.4 µA · id18 = 262.3 µA

    In saturation the current depends mainly on VGS: 25.0 µA for 0.9 V, 262 µA for 1.8 V. It still rises with VDS, which E2.4 measures.

Read the plot as a table
Drain-source voltage VDSid09id12id15id18
0 V0 A0 A0 A0 A
50 mV6.69 µA13.81 µA19.87 µA24.61 µA
200 mV18.92 µA46.53 µA70.82 µA90.23 µA
259.6 mV20.9 µA56.07 µA87.6 µA113 µA
400 mV22.55 µA71.17 µA119.6 µA159.5 µA
600 mV23.35 µA78.84 µA147.6 µA209.1 µA
740.3 mV23.69 µA80.74 µA155.9 µA231.8 µA
800 mV23.81 µA81.31 µA157.9 µA238.2 µA
1 V24.16 µA82.72 µA162 µA249.8 µA
1.2 V24.45 µA83.74 µA164.5 µA255.2 µA
1.4 V24.7 µA84.56 µA166.3 µA258.7 µA
1.6 V24.93 µA85.25 µA167.7 µA261.2 µA
1.7 V25.04 µA85.56 µA168.4 µA262.3 µA
1.8 V25.15 µA85.86 µA168.9 µA263.3 µA

Measured on ngspice 42, SKY130 tt (361 points). The knees are the model's own VDsat, read during the same run.

Below a certain gate voltage, the threshold voltage Vth, the device is in cutoff: almost no current flows. In triode the channel reaches from source to drain and the device behaves like a resistor. In saturation the channel pinches off near the drain once VDS passes VDsat, and the current is set mainly by the gate.

The textbook square law puts the knee at VDsat = VGS − Vth, the overdrive. With Vth = 0.556 V for this device (the lab below), that predicts knees at 0.34, 0.64, 0.94 and 1.24 V. The model puts them at 0.26, 0.44, 0.60 and 0.74 V. Even a 1 µm channel saturates early. The charge in the depletion layer under the channel pulls the knee in to about Vov/1.3 at low drive, and velocity saturation (E2.3) pulls it in further at high drive.

Two ways to extract a threshold

The channel does not switch on at a sharp voltage: the current grows smoothly over a few hundred millivolts. So Vth has to be defined by a method, and two are common. Both use a small VDS of 50 mV.

DC sweepOne device, two threshold methodsW/L = 2/0.15 µm at VDS = 50 mV. Above: ID and its tangent at the steepest point. Below: the same current on a log axis.
  • id
  • tangent
  • id_log
020 µA40 µA60 µA80 µAID, linear100 fA10 pA1 nA100 nA10 µAID, log scale00.2 V0.4 V0.6 V0.8 V1.0 V1.2 VGate-source voltage VGS100 nA × W/L = 1.33 µAconstant currenttangent at zero020 µA40 µA60 µA80 µAID, linear100 fA100 pA100 nA100 µAID, log scale00.5 V1.0 VGate-source voltage VGS100 nA × W/L = 1.33 µAconstant currenttangent at zero
  1. Gate-source voltage VGS 676.1 mVid = 1.333 µA · tangent = -11.22 µA · id_log = 1.333 µA

    Constant current: Vth is the VGS at which ID reaches 100 nA × W/L, here 1.33 µA. The log panel shows the crossing: Vth = 0.676 V.

  2. Gate-source voltage VGS 748.3 mVid = 4.804 µA · tangent = -2.183 pA · id_log = 4.804 µA

    Linear extrapolation: draw the tangent where the slope gm = dID/dVGS is largest and extend it to zero current. It lands at 0.748 V, and subtracting VDS/2 = 25 mV gives Vth = 0.723 V.

  3. Gate-source voltage VGS 938 mVid = 29.46 µA · tangent = 29.46 µA · id_log = 29.46 µA

    The tangent touches the curve at 0.938 V, where gm peaks at 155 µS. Above it, the slope falls as the field from the gate slows the carriers.

Read the plot as a table
Gate-source voltage VGSidtangentid_log
0 V130.4 fA-116.2 µA130.4 fA
200 mV12.56 pA-85.15 µA12.56 pA
400 mV2.098 nA-54.09 µA2.098 nA
600 mV268.1 nA-23.03 µA268.1 nA
676.1 mV1.333 µA-11.22 µA1.333 µA
748.3 mV4.804 µA-2.183 pA4.804 µA
800 mV9.718 µA8.032 µA9.718 µA
938 mV29.46 µA29.46 µA29.46 µA
1 V39.01 µA39.09 µA39.01 µA
1.2 V66.93 µA70.15 µA66.93 µA

Measured on ngspice 42, SKY130 tt (601 points). The tangent comes from the same run's gm.

The two methods differ by 47 mV on the same device. Neither is wrong, since they answer different questions, so a quoted threshold must name its method, its VDS, and the W and L it was measured on.

What moves the threshold

The lab extracts constant-current thresholds for a set of devices, then repeats with the drain raised and with the body lowered.

DeviceConditionMeasured Vth
NMOS 2/0.15VDS = 50 mVvtn_015 = 6.760510e-01
NMOS 2/0.5VDS = 50 mVvtn_05 = 5.941162e-01
NMOS 2/1VDS = 50 mVvtn_1 = 5.564164e-01
NMOS 2/4VDS = 50 mVvtn_4 = 5.244077e-01
NMOS 2/0.15VDS = 1.8 Vvtn_015_vd18 = 5.826013e-01
NMOS 2/1VDS = 1.8 Vvtn_1_vd18 = 5.349302e-01
NMOS 2/1VSB = 0.5 Vvtn_1_vsb05 = 6.603915e-01
NMOS 2/1VSB = 1 Vvtn_1_vsb1 = 7.450388e-01
PMOS 2/0.15|VDS| = 50 mVvtp_015 = 7.877186e-01
PMOS 2/1|VDS| = 50 mVvtp_1 = 9.905737e-01
PMOS 2/4|VDS| = 50 mVvtp_4 = 1.008237e+00

W/L in µm. PMOS values are |Vth|, the source-to-gate voltage. Measured on the lab netlist.

Length. The 0.15 µm NMOS has the highest threshold, 0.152 V above the 4 µm one, while many textbooks teach that short channels lose threshold. SKY130 adds extra doping at the ends of the channel, called halo implants, to limit leakage. In a short channel the halos fill more of the channel and raise Vth: the reverse short-channel effect.

Drain voltage. At VDS = 1.8 V the short device's threshold drops by 93 mV, the long one's by only 21 mV. The drain's field reaches the source end of a short channel and lowers the barrier electrons must cross: drain-induced barrier lowering (DIBL).

Body. Lowering the body 0.5 V below the source raises Vth by 104 mV; 1 V raises it by 189 mV. A source above its body needs more gate voltage to form the channel: the body effect. In a stack of NMOS devices, every one above the bottom pays it.

PMOS. The same rules hold with every sign flipped, but the SKY130 PMOS threshold is large: 0.79 V at L = 0.15 µm and 0.99 V at L = 1 µm. On a 1.8 V supply that leaves little overdrive, a budget every later circuit must respect.

Asked in interviews
  1. What is the threshold voltage, and how do you extract it?

    The gate voltage at which the channel forms. It is defined by a method: the VGS where ID = 100 nA × W/L, or the tangent at maximum gm extended to zero, minus VDS/2. On one SKY130 device they give 0.676 V and 0.723 V.

  2. Name the three regions and their conditions.

    Cutoff: VGS below Vth. Triode: VGS above Vth and VDS below VDsat, a gate-controlled resistor. Saturation: VDS above VDsat, current set by VGS. The square law says VDsat = VGS − Vth; short channels saturate earlier.

  3. Why can a short device have a higher threshold than a long one?

    Halo implants. Extra doping at the channel ends limits leakage, and in a short channel it covers more of the channel, raising Vth. This reverse short-channel effect competes with DIBL, which lowers Vth at high VDS.

Try it

Each WHEN i(VA)=1.333333u finds the gate voltage at which one device reaches 100 nA × W/L. alter changes a source between runs, so the same netlist measures the drain and body effects. Open Plots to see the four sweeps.

Try it: add an NMOS with W = 2 and L = 20 (the longest bin) and predict its threshold from the trend. Then set VDS to 0.9 V in the first run and see every short-channel threshold fall.

SPICEngspice 42 · simulator service
Edit the netlist, then run it
Expected measurements: vtn_015 = 6.760510e-01 vtn_05 = 5.941162e-01 vtn_1 = 5.564164e-01 vtn_4 = 5.244077e-01 vtn_7_8 = 5.022373e-01 vtn_7_015 = 6.481375e-01 vtp_015 = 7.877186e-01 vtp_1 = 9.905737e-01 vtp_4 = 1.008237e+00 vtn_015_vd18 = 5.826013e-01 vtn_1_vd18 = 5.349302e-01 vtn_1_vsb05 = 6.603915e-01 vtn_1_vsb1 = 7.450388e-01

Check yourself

Predict the outputThe lab's L = 1 µm NMOS now has its source lifted to 0.5 V above ground, with its body tied to its source. Gate and drain are driven from the source as before. Type the line meas prints (for example vt = 1.000000e+00).
* the L = 1 um device: source lifted to 0.5 V, body tied to the source
.lib sky130 tt
VS s 0 0.5
VGS g s 0
VDS d s 0.05
VC d d3 0
XN3 d3 g s s sky130_fd_pr__nfet_01v8 W=2 L=1
.control
dc VGS 0 1.8 0.002
meas dc vt WHEN i(VC)=200n
.endc
.end
ConceptBoth NMOS devices have W = 2 µm and VDS = 50 mV. Which has the higher constant-current threshold in SKY130, and why?
ConceptWhich statement of the saturation condition is wrong?
ConceptRead the output-curve figure. At VDS = 0.3 V, which curves are in triode?
ConceptWhat is drain-induced barrier lowering (DIBL)?

Remember

  • Cutoff, triode and saturation: below Vth almost no current; above it a gate-controlled resistor until VDS reaches VDsat; then a current set mainly by VGS.
  • A threshold needs a method: by constant current (100 nA × W/L) and by tangent extrapolation one SKY130 device gives 0.676 V and 0.723 V.
  • SKY130 short devices have the higher threshold (halo implants): 0.676 V at L = 0.15 µm against 0.524 V at 4 µm. DIBL lowers it at high VDS; body bias raises it.
  • The SKY130 PMOS threshold is 0.79 to 1.01 V, leaving little overdrive on a 1.8 V supply.