Short-Channel Effects a Designer Meets
Measure the short-channel effects that change a design. They are DIBL in mV/V, an ro that moves with VDS, and the currents that limit VDS.
You Will Be Able To
- Measure DIBL in mV/V from two gate sweeps and predict its effect on the off current.
- Read how ro changes with VDS and choose the drain voltage of a current source.
- Explain why hot carriers limit VDS, with the substrate current as evidence.
- State which leakage paths this SKY130 model includes and which it leaves out.
- Lesson
- E9.2
- Time
- 30 min
- Runs on
- ngspice
- Check
- 5 questions
A 1.5/0.15 µm switch passed its leakage test with 0.1 V across it: 64.6 fA. In the circuit it blocks 1.8 V and leaks 610 fA, 9.4 times more.
- 1.5/0.15, VDS = 0.1 V
- 1.5/0.15, VDS = 1.8 V
- 10/1, VDS = 0.1 V
- 10/1, VDS = 1.8 V
- Gate-source voltage VGS 0 V1.5/0.15, VDS = 0.1 V = 64.61 fA · 1.5/0.15, VDS = 1.8 V = 609.6 fA · 10/1, VDS = 0.1 V = 599.9 fA · 10/1, VDS = 1.8 V = 817.5 fA
Gate at 0 V. The short device leaks 64.6 fA at 0.1 V and 610 fA at 1.8 V.
- Gate-source voltage VGS 587 mV1.5/0.15, VDS = 0.1 V = 172.6 nA · 1.5/0.15, VDS = 1.8 V = 996.2 nA · 10/1, VDS = 0.1 V = 2.485 µA · 10/1, VDS = 1.8 V = 3.19 µA
At VDS = 1.8 V the short device reaches 1 µA at 587 mV.
- Gate-source voltage VGS 668 mV1.5/0.15, VDS = 0.1 V = 993.1 nA · 1.5/0.15, VDS = 1.8 V = 4.996 µA · 10/1, VDS = 0.1 V = 10.9 µA · 10/1, VDS = 1.8 V = 14.03 µA
At 0.1 V it needs 668 mV. The curve moved by 81 mV.
Read the plot as a table
| Gate-source voltage VGS | 1.5/0.15, VDS = 0.1 V | 1.5/0.15, VDS = 1.8 V | 10/1, VDS = 0.1 V | 10/1, VDS = 1.8 V |
|---|---|---|---|---|
| 0 V | 64.61 fA | 609.6 fA | 599.9 fA | 817.5 fA |
| 200 mV | 9.94 pA | 87.12 pA | 137.6 pA | 187.8 pA |
| 400 mV | 1.745 nA | 12.35 nA | 28.87 nA | 39.33 nA |
| 587 mV | 172.6 nA | 996.2 nA | 2.485 µA | 3.19 µA |
| 600 mV | 231.6 nA | 1.314 µA | 3.251 µA | 4.15 µA |
| 668 mV | 993.1 nA | 4.996 µA | 10.9 µA | 14.03 µA |
| 800 mV | 9.757 µA | 33.49 µA | 40.68 µA | 66.39 µA |
| 1 V | 49.29 µA | 141.9 µA | 92.46 µA | 222.6 µA |
The 10/1 device moves by 12 mV, and its leak rises 1.36 times.
The drain voltage moved the threshold of the short device. That effect has a name and a number.
Scaling, and Where It Stopped
Constant-field scaling shrinks every dimension and every voltage by the same factor k. The electric fields then stay the same.
| Quantity | Scales by |
|---|---|
| Length, width, oxide thickness | 1/k |
| Supply and threshold voltage | 1/k |
| Channel doping | k |
| Drain current, gate capacitance | 1/k |
| Gate delay | 1/k |
| Power per gate | 1/k2 |
| Power per area | 1 |
The rule that made each new process faster at the same heat per area.
SKY130 keeps the field nearly equal in its two oxides. The 1.8 V device has 4.148 nm of oxide, so 4.34 MV/cm. The 5 V device has 11.6 nm, so 4.31 MV/cm.
- Neutral
- Data
- DIBL
The drain voltage reaches across a short channel and lowers the barrier at the source.
- Impact ionization and GIDL
Both make current that leaves the drain for the body.
- Hot carriers
Fast carriers near the drain can enter the oxide and stay.
- Gate leakage
Carriers tunnel straight through a thin oxide.
Read the diagram as text
- Gate and oxide: controls the barrier
- Source: carriers wait at a barrier
- Channel: short: drain is near
- Drain: highest field
- Body: collects stray current
- Source → Channel: channel current
- Channel → Drain
- Drain → Body: to the body
- Drain → Gate and oxide: into the oxide
- Gate and oxide → Channel: through the oxide
The sections below measure each one.
Along the channel the rule broke. Lengths shrank faster than supplies, because a threshold cannot shrink without more leakage. The field along a short channel is therefore high, and the effects below follow.
DIBL as a Number
In a short channel the drain sits close to the source. A high drain voltage lowers the barrier that holds carriers back at the source. The gate then needs less voltage, so the threshold falls.
DIBL, drain-induced barrier lowering, is that fall per volt of VDS. The lab measures it at seven lengths.
| W/L (µm) | Vth at VDS = 0.1 V | Vth at VDS = 1.8 V | DIBL |
|---|---|---|---|
| 1.5/0.15 | 668.3 mV | 587.2 mV | 47.7 mV/V |
| 1.8/0.18 | 655.5 mV | 591.8 mV | 37.5 mV/V |
| 2.5/0.25 | 637.9 mV | 609.4 mV | 16.8 mV/V |
| 3.5/0.35 | 600.2 mV | 578.8 mV | 12.6 mV/V |
| 5/0.5 | 578.5 mV | 562.3 mV | 9.5 mV/V |
| 10/1 | 545.5 mV | 533.3 mV | 7.2 mV/V |
| 20/2 | 522.9 mV | 510.5 mV | 7.3 mV/V |
Vth is the VGS at ID = 1 µA. DIBL is the difference divided by 1.7 V.
DIBL falls from 48 mV/V at 0.15 µm to 10 mV/V at 0.5 µm. Beyond that, more length buys little.
Below the threshold, current depends exponentially on the gate voltage. A threshold that falls by tens of millivolts multiplies the off current, as the hook shows.
An Output Resistance That Moves with the Drain Voltage
- ro, 10/1
- ro, 1.5/0.15
- substrate current, 10/1
- Drain-source voltage VDS 400 mVro, 10/1 = 121.2 kΩ · ro, 1.5/0.15 = 71.68 kΩ · substrate current, 10/1 = 10.52 fA
Just into saturation, the 10/1 device has 121 kΩ.
- Drain-source voltage VDS 1 Vro, 10/1 = 304 kΩ · ro, 1.5/0.15 = 75.83 kΩ · substrate current, 10/1 = 206.7 fA
At 1 V it has 304 kΩ. The substrate current is 0.207 pA.
- Drain-source voltage VDS 1.8 Vro, 10/1 = 367.7 kΩ · ro, 1.5/0.15 = 76.84 kΩ · substrate current, 10/1 = 11.95 nA
Near 1.8 V ro peaks at 368 kΩ. The substrate current is now 11.95 nA.
- Drain-source voltage VDS 2.75 Vro, 10/1 = 192.5 kΩ · ro, 1.5/0.15 = 69.19 kΩ · substrate current, 10/1 = 933.4 nA
At 2.75 V ro is back to 192 kΩ, and 0.933 µA flows to the body.
Read the plot as a table
| Drain-source voltage VDS | ro, 10/1 | ro, 1.5/0.15 | substrate current, 10/1 |
|---|---|---|---|
| 400 mV | 121.2 kΩ | 71.68 kΩ | 10.52 fA |
| 500 mV | 171 kΩ | 73.26 kΩ | 10.62 fA |
| 1 V | 304 kΩ | 75.83 kΩ | 206.7 fA |
| 1.5 V | 358.5 kΩ | 76.78 kΩ | 884.6 pA |
| 1.8 V | 367.7 kΩ | 76.84 kΩ | 11.95 nA |
| 2 V | 355.9 kΩ | 76.41 kΩ | 42.76 nA |
| 2.5 V | 254.1 kΩ | 72.66 kΩ | 411.6 nA |
| 2.75 V | 192.5 kΩ | 69.19 kΩ | 933.4 nA |
The 1.5/0.15 device stays near 76.9 kΩ: DIBL holds its ro down.
The slope of ID against VDS has three causes, and each rules one range.
Low in saturation, channel-length modulation rules. It weakens as VDS grows, so ro rises.
In a short device DIBL adds a slope that does not weaken. Its ro stays low and flat.
At the top, fast carriers hit the lattice near the drain and free new carriers. This is impact ionization. The new current flows to the body, the drain current rises, and ro falls.
Hot Carriers and the Drain Voltage Limit
Carriers that gain high energy in the field near the drain are hot carriers. The substrate current counts them.
| Bias of the 10/1 device | Substrate current |
|---|---|
| VGS = 0.8 V, VDS = 1 V | 0.207 pA |
| VGS = 0.8 V, VDS = 1.8 V | 11.95 nA |
| VGS = 0 V, VDS = 1.8 V | 12.2 fA |
| VGS = 1.8 V, VDS = 0.2 V | 11 fA |
From the run behind the figure.
Hot carriers need two things at once: a channel current and a high drain voltage.
Some hot carriers enter the gate oxide and stay there. Over years this shifts the threshold and lowers the current. The foundry therefore limits VDS.
GIDL and Gate Leakage in This Model
GIDL is gate-induced drain leakage. A gate held far below the drain makes a strong field where the two overlap. Electrons tunnel from band to band in the drain under that overlap. The current flows from the drain to the body.
- 5 V NMOS, VDS = 5 V
- 5 V NMOS, VDS = 1.8 V
- 1.8 V NMOS, VDS = 1.8 V
- Gate-source voltage VGS -1.98 V5 V NMOS, VDS = 5 V = 719 fA · 5 V NMOS, VDS = 1.8 V = 11.82 fA · 1.8 V NMOS, VDS = 1.8 V = 11.8 fA
The 5 V device at VDS = 5 V leaks 719 fA. This is GIDL.
- Gate-source voltage VGS -1 V5 V NMOS, VDS = 5 V = 207.8 fA · 5 V NMOS, VDS = 1.8 V = 11.8 fA · 1.8 V NMOS, VDS = 1.8 V = 11.8 fA
With the gate at −1 V the same device leaks 208 fA.
- Gate-source voltage VGS -420 mV5 V NMOS, VDS = 5 V = 93.81 fA · 5 V NMOS, VDS = 1.8 V = 14.63 fA · 1.8 V NMOS, VDS = 1.8 V = 11.82 fA
Its minimum is 93.8 fA. Right of here the channel starts to conduct.
- Gate-source voltage VGS 300 mV5 V NMOS, VDS = 5 V = 975.3 pA · 5 V NMOS, VDS = 1.8 V = 329.8 pA · 1.8 V NMOS, VDS = 1.8 V = 1.046 nA
Subthreshold current now hides everything else.
Read the plot as a table
| Gate-source voltage VGS | 5 V NMOS, VDS = 5 V | 5 V NMOS, VDS = 1.8 V | 1.8 V NMOS, VDS = 1.8 V |
|---|---|---|---|
| -2 V | 735.1 fA | 11.83 fA | 11.8 fA |
| -1.98 V | 719 fA | 11.82 fA | 11.8 fA |
| -1.5 V | 408.2 fA | 11.8 fA | 11.8 fA |
| -1 V | 207.8 fA | 11.8 fA | 11.8 fA |
| -500 mV | 99.34 fA | 12.57 fA | 11.8 fA |
| -420 mV | 93.81 fA | 14.63 fA | 11.82 fA |
| 0 V | 7.609 pA | 2.57 pA | 609.6 fA |
| 300 mV | 975.3 pA | 329.8 pA | 1.046 nA |
The 1.8 V device stays at 11.8 fA down to −2 V: no GIDL in its card.
The 5 V card has a GIDL parameter, agidl = 5.06e-11. The 1.8 V card sets agidl to 0, so its curve is flat.
Gate leakage is tunnelling through the oxide itself. Both cards switch that model off, and ngspice prints exactly 0 A of gate current.
Asked in interviews
What is DIBL, and how do you measure it?
A high drain voltage lowers the source barrier of a short device, so the threshold falls. Measure the threshold at a low and a high VDS and divide the difference by the voltage step. Here it is 48 mV/V at 0.15 µm and 7 mV/V at 1 µm.
Why does ro depend on VDS?
Channel-length modulation weakens as VDS rises, so ro rises. In short devices DIBL keeps ro low. At high VDS impact ionization adds current to the body and ro falls again.
What are hot carriers, and why do they limit VDS?
Carriers that gain high energy in the field near the drain. Some enter the gate oxide and shift the threshold over time. The substrate current is their signature, so designers keep VDS below the rating while current flows.
Try It
All seven devices have W/L = 10, so one current, 1 µA, marks the threshold of each. Parts 1 and 2 read it at VDS = 1.8 V and 0.1 V.
DIBL is the difference divided by 1.7 V. For the shortest device that is (668.3 − 587.2) mV / 1.7 V = 47.7 mV/V.
Part 3 holds the gate at 0.8 V and sweeps the drain. deriv() gives gds. At 0.2 V the 10/1 device is still close to triode, with 69.6 µS. From 0.6 V to 1.75 V it falls from 4.76 µS to 2.72 µS. The short device stops near 13 µS.
Predict before each change. Set every L twice as long and every W twice as wide: which DIBL values move most? Set VG in part 3 to 1.2: does gds of the 10/1 device still fall by the same factor?
Run the netlist to draw its plots here.
vhi_a = 5.871763e-01 vhi_b = 5.918051e-01 vhi_c = 6.094255e-01 vhi_d = 5.787631e-01 vhi_e = 5.622842e-01 vhi_f = 5.333321e-01 vhi_g = 5.104844e-01 vlo_a = 6.683369e-01 vlo_b = 6.555436e-01 vlo_c = 6.379132e-01 vlo_d = 6.002439e-01 vlo_e = 5.785132e-01 vlo_f = 5.455362e-01 vlo_g = 5.228670e-01 gds_a_02 = 1.677519e-05 gds_f_02 = 6.961320e-05 gds_a_06 = 1.348092e-05 gds_f_06 = 4.758439e-06 gds_a_10 = 1.318704e-05 gds_f_10 = 3.289560e-06 gds_a_14 = 1.304657e-05 gds_f_14 = 2.849619e-06 gds_a_175 = 1.300685e-05 gds_f_175 = 2.717250e-06 id_a = 2.303652e-05 id_f = 6.407438e-05Check Yourself
Common Mistakes
Wrong Characterize the leakage of a short switch at a small VDS.
DIBL lowers the threshold at a high drain voltage, so the off current there is several times larger.
Fix Measure the off current at the largest VDS the switch must block, and at the hottest temperature.
Wrong Quote one ro for a device.
ro moves with VDS: the 10/1 device here goes from 121 kΩ to 368 kΩ.
Fix Read gds at the drain voltage the device will have in the circuit.
dc VD 0 1.8 0.01
let gds=deriv(i(VIf))
meas dc gds FIND gds AT=0.9dc VD 0 1.8 0.01
let gds=deriv(i(VIf))
* the drain of this device sits at 0.4 V
meas dc gds FIND gds AT=0.4Wrong Read zero gate current in the simulation as proof that the oxide does not leak.
The SKY130 cards switch the gate-current model off, so the simulator has nothing to compute.
Fix Check which effects the card includes before you trust a missing current.
Remember
- DIBL is the threshold drop per volt of VDS. Here it is 48 mV/V at 0.15 µm and 7 mV/V at 1 µm.
- A lower threshold multiplies the off current, so test leakage at the highest drain voltage.
- ro rises with VDS while channel-length modulation weakens, and falls once impact ionization starts.
- Hot carriers need current and a high VDS together. The substrate current is their signature.
- This model has GIDL only in the 5 V device and no gate current in either. A missing effect is not proof.