Reading a Model Card and an Operating Point

Read the model card behind a SKY130 transistor and the operating point ngspice prints, and place the square-law hand model inside them.

You Will Be Able To

  • Print a device's model card with showmod and its operating point with show.
  • Name the card that a W and an L select, and predict when a new size selects another.
  • Separate a threshold, mobility, velocity or DIBL change with readings at two drain voltages.
  • Check a square-law estimate against the printed vth, vdsat and gm.
Lesson
E9.1
Time
30 min
Runs on
ngspice
Check
5 questions

A designer needs 100 µA from a 1/0.15 µm NMOS. The model card gives a threshold vth0 of 0.562 V and a mobility u0, and the square law turns them into VGS = 0.892 V. The simulator gives 32.3 µA there.

DC sweepCard Values in the Square Law, Against the Full ModelNMOS W/L = 1/0.15 µm at VDS = 1 V.
  • full model
  • square law from the card
00.2 mA0.4 mA0.6 mA0.8 mA1.0 mA1.2 mA1.4 mA1.6 mADrain current ID00.2 V0.4 V0.6 V0.8 V1.0 V1.2 V1.4 V1.6 V1.8 VGate-source voltage VGShand bias, 0.892 V00.5 mA1.0 mA1.5 mADrain current ID00.5 V1.0 V1.5 VGate-source voltage VGShand bias, 0.892 V
  1. Gate-source voltage VGS 892 mVfull model = 32.27 µA · square law from the card = 100.1 µA

    The hand model promises 100.1 µA. The full model gives 32.27 µA.

  2. Gate-source voltage VGS 1.085 Vfull model = 99.91 µA · square law from the card = 251.3 µA

    The device needs 1.085 V for 100 µA.

  3. Gate-source voltage VGS 1.8 Vfull model = 432.3 µA · square law from the card = 1.407 mA

    At 1.8 V the hand model says 1.41 mA. The device carries 432 µA.

Read the plot as a table
Gate-source voltage VGSfull modelsquare law from the card
0 V1.115 pA0 A
200 mV19.88 pA0 A
400 mV3.287 nA0 A
600 mV419.4 nA1.328 µA
800 mV12.71 µA52.08 µA
892 mV32.27 µA100.1 µA
1 V66.95 µA176.3 µA
1.085 V99.91 µA251.3 µA
1.2 V148.9 µA373.8 µA
1.4 V240.7 µA644.9 µA
1.6 V336 µA989.3 µA
1.8 V432.3 µA1.407 mA

One run. The hand curve is a behavioural source in the same netlist.

The card is not wrong: it holds far more than the two numbers the square law uses.

A Compact Model Is Equations Plus a Card

What the Simulator Does for Every TransistorArrows show what feeds what.
  • File
  • Constraint
  • Source
  • Model
  • Result
  • Measure
  • Data
What the simulator does for every transistorArrows show what feeds what. file Model card fitted numbers constraint Size W and L pick a card source Bias VGS, VDS, VBS, T model BSIM4 equations the same for everycard result Currents andcharges used to solve thecircuit measure Operating point vth, vdsat, gm, gds,cgg
  1. file Model cardfitted numbers
    • BSIM4 equations
  2. constraint SizeW and L pick a card
    • BSIM4 equations
  3. source BiasVGS, VDS, VBS, T
    • BSIM4 equations
  4. model BSIM4 equationsthe same for every card
    • Currents and charges
    • Operating point
  5. result Currents and chargesused to solve the circuit
  6. measure Operating pointvth, vdsat, gm, gds, cgg
  1. The card

    The foundry measured real transistors and fitted these numbers. showmod prints them.

  2. The equations

    The equations turn the card, the size and the bias into currents and charges.

  3. The printout

    The equations also give small-signal values at the bias. show prints them.

Read the diagram as text
  • Model card: fitted numbers
  • Size: W and L pick a card
  • Bias: VGS, VDS, VBS, T
  • BSIM4 equations: the same for every card
  • Currents and charges: used to solve the circuit
  • Operating point: vth, vdsat, gm, gds, cgg
  • Model card → BSIM4 equations
  • Size → BSIM4 equations
  • Bias → BSIM4 equations
  • BSIM4 equations → Currents and charges
  • BSIM4 equations → Operating point

The lab runs both commands.

A compact model is a fixed set of equations for a transistor's currents and charges. A model card is the list of numbers fitted to those equations. SKY130 cards use the BSIM4 equations, version 4.5.

ParameterWhat it setsCard of the 1/0.15 device
vth0Threshold at zero body bias0.5617 V
u0Mobility at low field0.03307 m2/(V·s)
vsatTop speed of the carriers165412 m/s
eta0DIBL: threshold drop per volt of VDS−0.7482
toxeOxide thickness, so Cox4.148 nm

Five of the 350 values in this card. From showmod in the lab.

DIBL is drain-induced barrier lowering: a high drain voltage lowers the threshold of a short device. E9.2 measures it.

One Parameter, One Signature

altermod changes one card parameter between runs. The figure shows the card as shipped and three altered copies.

DC sweepEach Parameter Bends the Curve Its Own WayID against VGS at VDS = 1.8 V.
  • card as shipped
  • vth0 + 0.1 V
  • u0 − 30 %
  • vsat − 30 %
0100 µA200 µA300 µA400 µA500 µADrain current ID00.2 V0.4 V0.6 V0.8 V1.0 V1.2 V1.4 V1.6 V1.8 VGate-source voltage VGS0100 µA200 µA300 µA400 µA500 µADrain current ID00.5 V1.0 V1.5 VGate-source voltage VGS
  1. Gate-source voltage VGS 900 mVcard as shipped = 46.8 µA · vth0 + 0.1 V = 18.47 µA · u0 − 30 % = 31.62 µA · vsat − 30 % = 43.62 µA

    Threshold: vth0 + 0.1 V slides the whole curve right by 100.7 mV.

  2. Gate-source voltage VGS 1.3 Vcard as shipped = 224.6 µA · vth0 + 0.1 V = 172.6 µA · u0 − 30 % = 170.1 µA · vsat − 30 % = 193.3 µA

    Mobility: u0 scales the current everywhere above the threshold.

  3. Gate-source voltage VGS 1.8 Vcard as shipped = 475.6 µA · vth0 + 0.1 V = 416.1 µA · u0 − 30 % = 385 µA · vsat − 30 % = 394.2 µA

    Velocity: vsat costs 17.1 % at full drive and almost nothing near the threshold.

Read the plot as a table
Gate-source voltage VGScard as shippedvth0 + 0.1 Vu0 − 30 %vsat − 30 %
0 V2.046 pA1.828 pA1.959 pA2.042 pA
200 mV39.98 pA4.769 pA25.82 pA39.35 pA
400 mV5.999 nA477 pA3.774 nA5.9 nA
600 mV697 nA69.35 nA439.1 nA684.7 nA
800 mV18.77 µA4.639 µA12.23 µA17.99 µA
900 mV46.8 µA18.47 µA31.62 µA43.62 µA
1 V85.25 µA45.98 µA59.69 µA77.37 µA
1.2 V176.1 µA126.7 µA130.5 µA153.7 µA
1.3 V224.6 µA172.6 µA170.1 µA193.3 µA
1.4 V274.1 µA220 µA211.4 µA233.4 µA
1.6 V374.5 µA317.4 µA297.2 µA313.9 µA
1.8 V475.6 µA416.1 µA385 µA394.2 µA

The run repeats the sweep at VDS = 0.1 V for the table.

ChangeVth shift, VDS = 0.1 VVth shift, VDS = 1.8 VID at VGS = 1.8 V, VDS = 0.1 VID at VGS = VDS = 1.8 V
vth0 + 0.1 V100.8 mV100.7 mV−9.9 %−12.5 %
u0 − 30 %23.9 mV22.0 mV−30.1 %−19.1 %
vsat − 30 %0.6 mV0.8 mV−1.9 %−17.1 %
eta0 + 0.05−1.2 mV−20.1 mV0.1 %2.4 %
toxe + 10 %85.7 mV79.2 mV−10.8 %−14.6 %

Vth here is the VGS at ID = 667 nA, the constant-current method of E2.2. The shipped card reads 680 mV and 597.9 mV.

A threshold change shifts the curve equally at both drain voltages. A DIBL change shifts it only at high VDS. Mobility shows most in the triode current, and velocity in the saturated current.

A single reading of ID cannot tell these apart. Readings at two drain voltages can.

Bins: 48 Cards Under One Name

No single card fits every size. The foundry fits one card for each range of W and L, called a bin. This NMOS has 48 cards, and ngspice picks one from the W and L of each instance.

W/L (µm)Card selectedL rangeW rangevth0 in the cardvth printed
1/0.15nshort_model.400.15 to 0.18 µm0.55 to 1 µm0.562 V0.77 V
1/1nshort_model.370.5 to 1 µm0.55 to 1 µm0.642 V0.616 V
10/8nshort_model.24 to 8 µm7 to 1010 µm0.513 V0.526 V
2/1nshort_model.290.5 to 1 µm1 to 3 µm0.566 V0.606 V

From the lab, at VGS = VDS = 1 V.

Inside a bin, each parameter still changes with size. The card stores a base value P0 and three size terms: P = P0 + PL/Leff + PW/Weff + PP/(Leff·Weff).

So vth0 and eta0 in a card are base values. For the 1/0.15 device, eta0 is −0.748, and a negative DIBL would make no physical sense. The size terms make up the difference.

Sizes on two sides of a bin edge use different cards. The 2/1 device carries 2.157 times the current of the 1/1 device.

The Operating Point, Line by Line

The Printout as a Small-Signal ModelThe 1/0.15 device at VGS = VDS = 1 V.
The printout as a small-signal modelgateCgg​0.797 fFgm·vgs377.6 µSro​41.27 kΩdrain
  1. gate · portterminal → n1
  2. C_gg · capacitor0.797 fFpin 1 → n1 · pin 2 → 0
  3. gm·vgs · current source377.6 µS+ → n2 · − → 0
  4. r_o · resistor41.27 kΩpin 1 → n2 · pin 2 → 0
  5. drain · portterminal → n2
  6. gnd · groundgnd → 0
  1. gm

    gm is the change of drain current per volt of gate voltage.

    gm·vgs

  2. gds

    gds is the change of drain current per volt of drain voltage. Its inverse is ro.

  3. cgg

    cgg is the total capacitance seen at the gate.

Read the circuit as text
  • gate port: terminal → n1
  • C_gg capacitor (0.797 fF): pin 1 → n1, pin 2 → 0
  • gm·vgs current source (377.6 µS): + → n2, − → 0
  • r_o resistor (41.27 kΩ): pin 1 → n2, pin 2 → 0
  • drain port: terminal → n2

E3.1 splits Cgg into its parts.

Line of show1/0.151/110/8
vth0.77 V0.616 V0.526 V
vdsat0.169 V0.333 V0.397 V
id66.95 µA18.79 µA32.58 µA
gm377.6 µS86.21 µS130.1 µS
gds24.23 µS1.241 µS0.4366 µS
cgg0.7974 fF5.595 fF465.2 fF

All three at VGS = VDS = 1 V.

vth is the model's own threshold at this bias, with every size and bias term included. It is not an extracted threshold. The 1/0.15 device prints 0.77 V, while the constant-current method reads 0.622 V at the same VDS.

vdsat is the drain voltage at which the model enters saturation. Compare it with VDS to name the region.

Where the Square Law Sits

The square law keeps three numbers: one threshold, one current factor µ·Cox·W/L, and one λ. Feed it the printed values and compare.

At VGS = VDS = 1 V10/8 printed10/8 square law1/0.15 printed1/0.15 square law
ID, from the card's vth0 and u032.58 µA39.74 µA66.95 µA176.3 µA
gm = 2ID/(VGS − vth)130.1 µS137.5 µS377.6 µS583.1 µS
VDsat = VGS − vth0.397 V0.474 V0.169 V0.23 V

Cox = 3.9·ε0/toxe = 8.325 fF/µm2.

For the long, wide device the hand current is 22 % high and the hand gm is 6 % high. For the short device the hand current is 2.6 times too large.

The full model adds what the square law leaves out. Mobility falls with the gate field, and carriers reach a velocity limit. The threshold moves with size and drain voltage.

Asked in interviews
  1. What is a compact model, and what is a model card?

    A compact model is a fixed set of equations for a device's currents and charges, such as BSIM4. A model card is the list of numbers a foundry fitted to those equations, for one device type and one size range.

  2. What is binning?

    One card cannot fit every W and L, so the foundry fits a card for each range of sizes. The simulator picks the card from the instance size. Here a 1/1 device and a 2/1 device use different cards, and the current rises 2.157 times.

  3. Which operating-point values do you read first?

    Read vdsat against VDS for the region. Then read gm and gds for the gain, and cgg for the load on the stage before. Read vth only as the model's internal value.

Try It

The save lines keep six operating-point values of each device. The three-point sweep then gives meas a middle point to read.

Part 2 reads the card of each device. @xm1:nshort_model.40[vth0] is one parameter of the card that XM1 selected. The card names are fixed in these lines, so they match these four sizes only.

Part 3 prints the same facts in the log, as a designer usually reads them. showmod also names the card of each device.

Predict before each change. Set XM2 to W=1.2: which card does showmod name now? Set VD to 0.2: which devices still have VDS above vdsat?

SPICEngspice 42 · simulator service
Edit the netlist, then run it
Expected measurements: vth_1 = 7.703535e-01 vdsat_1 = 1.692331e-01 id_1 = 6.695303e-05 gm_1 = 3.776094e-04 gds_1 = 2.422936e-05 cgg_1 = 7.974000e-16 vth_2 = 6.155600e-01 vdsat_2 = 3.326711e-01 id_2 = 1.879459e-05 gm_2 = 8.620974e-05 gds_2 = 1.241425e-06 cgg_2 = 5.594736e-15 vth_3 = 5.260026e-01 vdsat_3 = 3.973791e-01 id_3 = 3.257909e-05 gm_3 = 1.300898e-04 gds_3 = 4.366313e-07 cgg_3 = 4.652203e-13 vth_4 = 6.057846e-01 id_4 = 4.054021e-05 lmin_1 = 1.500000e-07 lmax_1 = 1.800000e-07 wmin_1 = 5.500000e-07 wmax_1 = 1.000000e-06 vth0_1 = 5.617438e-01 u0_1 = 3.307353e-02 lmin_2 = 5.000000e-07 lmax_2 = 1.000000e-06 wmin_2 = 5.500000e-07 wmax_2 = 1.000000e-06 vth0_2 = 6.421817e-01 u0_2 = 2.380725e-02 lmin_3 = 4.000000e-06 lmax_3 = 8.000000e-06 wmin_3 = 7.000000e-06 wmax_3 = 1.010000e-03 vth0_3 = 5.128170e-01 u0_3 = 3.217780e-02 lmin_4 = 5.000000e-07 lmax_4 = 1.000000e-06 wmin_4 = 1.000000e-06 wmax_4 = 3.000000e-06 vth0_4 = 5.660596e-01 u0_4 = 2.799000e-02

Check Yourself

Predict the outputA designer builds W = 2 µm from two fingers of 1/1 µm, at the lab's bias. Type the line meas prints for the current in one finger (format: id_finger = 1.234567e-05).
* W = 2 um built from two fingers of W/L = 1/1, at the lab's bias
.lib sky130 tt
VG g 0 1.0
VD d 0 1.0
VA d da 0
XMA da g 0 0 sky130_fd_pr__nfet_01v8 W=1 L=1
XMB d g 0 0 sky130_fd_pr__nfet_01v8 W=1 L=1
.control
dc VG 0.9 1.1 0.1
meas dc id_finger FIND i(VA) AT=1
.endc
.end
ConceptRead the figure with the four curves. Which change leaves the current near VGS = 0.9 V almost unchanged, and lowers it clearly at 1.8 V?
ConceptA measured device has the threshold of the model at VDS = 0.1 V. At VDS = 1.8 V its threshold reads 20 mV lower than the model. What differs?
ConceptThe card of the 1/0.15 device lists vth0 = 0.5617 V. The operating point prints vth = 0.770 V. Why do they differ?
ConceptThe 10/8 device prints cgg = 465 fF in saturation. W·L·Cox is 666 fF. What explains most of the gap?

Common Mistakes

Wrong Take vth0 from the card as the threshold of the device.

The card value is a base term, and size and bias terms move the real threshold.

Fix Read vth at your bias with show. The 1/0.15 device prints 0.77 V against 0.562 V in the card.

Wrong Put two parameters on one altermod line.

In ngspice 42 that line changes only its first parameter, and it gives no warning.

Fix Write one altermod line for each parameter, then run again.

Before
altermod xm1:nshort_model.40 vth0=0.60 u0=0.04
After
altermod xm1:nshort_model.40 vth0=0.60
altermod xm1:nshort_model.40 u0=0.04

Wrong Diagnose a device from one drain current at one bias.

A threshold, mobility, velocity or DIBL change can each give the same current at one point.

Fix Read the threshold at low and high VDS, then the triode and the saturated current.

Remember

  • A compact model is a set of equations. A model card is the foundry's fitted numbers for one range of sizes.
  • W and L select the card. A size across a bin edge uses another card with other numbers.
  • Card values such as vth0 are base terms. The printed vth, vdsat, gm, gds and cgg describe the device at its bias.
  • Two drain voltages separate a threshold change from a DIBL change. Triode and saturated currents separate mobility from velocity.
  • The square law is close for a long, wide device and wrong by 2.6 times for the 1/0.15 device here.

Practice

Report a problem with this page