- 1. Transistor Size
- 2. Buffer
- 3. Body Effect Factor
- 4. MOS Capacitances
- 5. NOR Gate
- 6. Inverter Noise Margin

*In the circuit of Figure below, design the widths of the pull-down transistors so that V _{OL}=0.1 V. (All transistors are minimum size, L = 0.1 μm. Explain the results.*

Since the diagrams don’t show how the transistor bulk are connected/biased, we can assume that there is no body-bias; and V_{T} = V_{T0} = 0.4 V.

We first take a look at the extreme ends of operation. When Vin is high, VOH is VDD because the MOS is in cut-off. Thus no current flows through the resistor and the voltage drop across the resistor is 0.

Since our parameter of interest is V_{OL} and these are inverters, we are interested when the input voltage is HIGH.

The resistive load transistor operates in the **linear** region. So we equate the current in the resistor to the current through the NMOS (I_{DS}):

We substitute:

- V
_{GS}is V_{in}which is also V_{DD}. - V
_{DS}is V_{out}which is also V_{OL}if the input voltage V_{in}is V_{DD}. - V
_{DD}is 1.2 V. - V
_{OL}is 0.1 V. - R
_{L}is 10 kΩ. - L is 0.1 μm.
- μ
_{n}is the mobility: 270 cm^{2}V^{-1}s^{-1}. - C
_{ox}is the oxide capacitance: 1.6×10^{-6}F/cm^{2}. - E
_{C}is the critical field: 6 V/μm. - V
_{T}is 0.4 V.

Isolate and solving for W:

Note that the centimeter (cm) cancels out when we multiply μ_{n} and C_{ox}. So we do the rest of the calculations with μm as the unit for length.

The width of the NMOS transistor for the resistive-load inverter is **0.4 μm**.

Let’s denote the top transistor as *L* for *load*, and the bottom transistor as *I* for *inverting*. The top NMOS gate voltage is tied with V_{DD}, so it is always operating in saturation. The bottom transistor is still operating in linear mode.

Again, we start by equating the current of the top transistor with the bottom transistor: I_{DSL} = I _{DSI}.

Where we substitute:

- V
_{GSI}is just V_{in}. - V
_{DSI}is V_{out}. - V
_{T}is 0.4 V since we’re not assuming any body-bias effects. - V
_{GSL}is V_{DD}- V_{out}.

Isolate for W_{I}:

Also note that in order to use v_{sat} = 8×10^{6} cm/s, we need to convert this to μm/s, so we multiply the RHS of the equation by 1×10^{-4}.

In this case the gate voltage for the load transistor is 0.4 V higher than V_{DD} so the maximum output can be V_{DD}. But because the gate voltage is also higher, the top transistor is less-saturated.

Using the exact same equation as above, we use 1.6 - V_{out} instead for V_{GSL}. But V_{in} (for the bottom NMOS) remains at V_{DD} of 1.2 V.

The resistive-load inverter is able to have a output voltage all the way up to V_{DD} which is good. But the resistor is cumbersome in digital design, and often means slower and bigger circuit. The size of the transistor thus is also relatively large at **0.4 μm**.

Using the saturated-enhancement-load inverter, we no longer need the resistor. However, the maximum output voltage (high) is limited to V_{DD} - V_{T}. The trade-off is compensated with the small transistor size: at only **0.2 μm**.

Lastly, we bias the gate voltage of the load NMOS exactly V_{T} higher to obtain a maximum high output voltage of V_{DD}. But the result is we use a substantially larger transistor: at **0.3 μm**.

The design presented in the diagram is a **buffer** or voltage follower or something like the *Redstone repeater* from *Minecraft*.

Output swing: because the PMOS is the pull-down device, V_{OL} is at least V_{TP}, the threshold voltage from the PMOS. Similarly, because the NMOS is used as a pull-up device, V_{OH} is at most V_{DD} - V_{TN}.

- Starting from V
_{in}of 0.0 V and slowly increasing it to V_{T}. The output won’t budget because the PMOS remains on while the NMOS remains off. - After V
_{in}= V_{Tn}+ V_{TP}, the driving NMOS turns on and start to pull output up. The output voltage will keep increasing as we increase the input voltage, all the way to V_{DD}. By then, the output voltage is capped at V_{DD}- V_{TN}. V_{DD}- V_{TP}is where the PMOS fully turns off, which is close to where V_{OH}is. - Now let’s go from input high to input low. The output won’t budge until V
_{in}drops to V_{DD}- V_{TP}. because the pull-down PMOS is off. - At V
_{in}= V_{DD}- V_{TP}- V_{TN}and below, the PMOS gets stronger and pulls the output voltage back down to V_{TP}.

Notice that positive edge characteristics is different from negative edge, therefore we have hysteresis. It looks like this:

Theoretically, the gain should be close to 1 as output voltage follows input voltage. However, this is not a valid gate because it fails the following requirements:

- :x: High gain region is not between low gain regions.
- :x: Gain is higher than 1 for high gain region.
- :x: Low output is not below V
_{IL}. - :x: High output is not above V
_{IL}.

In additional to lack of noise-rejection, this design also lack regenerative properties.

Below is the schematic. Since the bulk connection is not specified from the diagram. The bulk of the NMOS is connected to ground, and the bulk of the PMOS is connected to V_{DD}.

Running the DC sweep on V_{in} from 0V to 1.2V (V_{DD}) gets us the follow VTC plot. I also took the derivative of the green curve (VTC curve) so we can see that there exists hysteresis more clearly. The plot is a bit different from what I expected.

First model on CAD:

To take multiple samples, I fix V_{2} to some value (0V, 0.2V, 0.4V, etc.). And then perform a DC sweep for V_{1} from 0 V to 1.2 V. For each trial, V_{T} is where the current I_{DS} begin to rise: see below sample calculation and plot. I’m using 1 μA as a threshold to determine the V_{T}.

Here’s a table of data for V_{T} using different V_{2} values:

V_{2} [V] |
V_{T} [V] |
---|---|

0.0 | 0.419 |

0.2 | 0.645 |

0.4 | 0.870 |

0.6 | 1.092 |

0.8 | 1.313 |

Using the data collected above, we can compute for the body-effect coefficient $\gamma$:

Where V_{T0} is 0.4 V, V_{SB} is just V_{2}, and 2φ_{F} is 0.88V.

V_{T0} is when the second term is zero. Since $\gamma$ and $2\phi_F$ is non-zero, the only case is when V_{SB} is 0. Which gives us **V _{T0}=0.419 V**.

I used *Excel* to compute the other data where V_{SB}=V_{2} is not zero and using the formula:

V_{2} [V] |
V_{T} [V] |
Body-Effect Coefficient [V^{½}] |
---|---|---|

0.2 | 0.645 | 0.12 |

0.4 | 0.870 | 0.23 |

0.6 | 1.092 | 0.32 |

0.8 | 1.313 | 0.41 |

The **body-efficient coefficient** $\gamma$ varies with the voltage across source and bulk and it ranges between **0.12 and 0.41**.

*Consider the layout in the figure below implemented in a 180nm technology. Assume that the transistor has W=900nm, L=180m, and a source/drain dimension Y=800nm and a lateral diffusion of 22nm. Let tox = 40 Å.*

*Compute the worst case gate capacitance per unit width, C _{g} in units of fF/μm. Estimate C_{GS}, C_{GD} and C_{GB} in linear, saturation, and cutoff, including overlap effects.*

Let’s first find the oxide capacitance, which is a function of the thickness of the oxide layer.

Where $\epsilon_{ox}$ is the silicon dioxide relative permissibility multiplied by the permissibility of free-space: $3.9\epsilon_0$. And the thickness $t_{ox}$ is 40 angstroms, or 40×10^{-10} m.

Plugging in the values, we get values $C_{ox}=8.63$ F/μm^{2}.

Then the gate capacitance is simply multiplication by the length L:

Since the thickness of the gate, T_{poly} is not given, we assume that the fringe capacitance is negligible (C_{f} = 0).

The overlap capacitance for each overlap is given by the oxide capacitance multiplied by how much is overlapped (lateral diffusion L_{D}=22 nm).

The total gate capacitance (which need us to multiply the numbers we got earlier with the width W) is

The total overlap capacitance for each overlap is

The capacitances broken down into the three parts: gate-source, gate-drain, and gate-bulk is as follows:

Cutoff | [fF] | Saturation | [fF] | Linear | [fF] | |
---|---|---|---|---|---|---|

V_{GS} |
C_{OL} |
0.17 | ⅔C_{G}+C_{OL} |
1.10 | ½C_{G}+C_{OL} |
0.87 |

V_{GD} |
C_{OL} |
0.17 | C_{OL} |
0.17 | ½C_{G}+C_{OL} |
0.87 |

V_{GB} |
C_{G} |
1.40 | - | 0.00 | - | 0.00 |

First, we calculate the voltage asymptote: φ_{B}.

Next, the zero-bias junction capacitance, where ε_{si} is the pure silicon permissibility at 11.7ε_{0}:

Lastly, to get C_{j}, we multiply the per-area capacitance with the area: which is the bottom plate and the side walls.

To calculate the junction capacitance, the formula is

Where A is the area, V_{J} is the voltage across the junction, and m is ½.

When V_{D} is 1.8V the voltage across the junction is the voltage of the bulk subtract the drain voltage:

Plugging in, we get

When V_{D} is 0, the junction voltage is 0. So junction capacitance is simply

*Calculate Vs of 2-input NOR gates when one input is switching and with both inputs tied together. The device sizes are W _{P} = 24λ and W_{N} = 6λ. Use Cadence to find results when switching only input A, AB together. Results for two inputs switched separately vary slightly. Explain the discrepancy between theory and simulation.*

First, I modelled the NOR gate in CAD. Note that I left the width property of the PMOS and NMOS as variable `WP`

and `WN`

. Here, we are using λ = 100nm.

Next, I created the test-bench for just switching input A:

I setup the DC sweep analysis for *Vin* from 0V to 1.2V (V_{DD}). Here is the voltage transfer characteristics (VTC):

The switching voltage **V _{S} = 0.642V**.

Now I modify the test-bench such that the input A and B are tied up together:

And here is the corresponding VTC plot:

Now the switching voltage **V _{S} = 0.588V** has shifted to the left.

“How much stronger the NMOS is to the PMOS” can be described by the factor $\chi$:

And the switching voltage is given by:

So let’s find the threshold voltage first. Since the bulk of the PMOS are connected to V_{DD} and the bulk of the NMOS are connected to ground, V_{SB} = 0 for both, and therefore V_{T} is 0.4V.

The two extreme cases are: if only one input is switching, and if all inputs are tied as one. Any behavior of the gate (in terms of the VTC) is bracketed between these two cases.

**Case 1: only A is switching**:

As per standard CMOS design, for a NOR gate, the PMOS would be 4 times larger than the NMOS. However, because the MOS connected to B never switches and is grounded — the B-PMOS is a short, and B-NMOS is an open circuit.

The resulting (simplified) circuit is a single-input inverter:

Note that in the diagram, I used “W” to show relative size of the NMOS and PMOS. To fit the problem description, W is equivalent to 6λ.

Compared to a standard inverter where the PMOS is suppose to have the double the width that of NMOS, we have 4 times the width. Meaning that the pull-up PMOS is much “stronger”. In other words, $\chi$ is smaller. Thus, we expect the VTC to shift to the right →.

Switching voltage:

**Case 2: all inputs tied:**

Shorting all the inputs together and using the same simplification used for case 1, we get the following equivalent single-input inverter (adding resistances: 4W and 4W in series makes 2W; W and W in parallel makes 2W):

Now compared to the standard inverter, NMOS is now much stronger than PMOS (since it should’ve been half the size). So $\chi$ is larger, and we expect the VTC to shift to the left.

Switching voltage:

The calculated switching voltages are smaller compared to the voltage measured from the simulation. I suspect it’s because there exists unaccounted body-effect. This could change the threshold voltage V_{T} and thus change the switching voltage V_{s}. V_{T} of 0.56 V would gives us closer number to the one obtained from simulation.

*Using the following schematic and Cadence simulations find NM _{H} and NM_{L}*.

**Note:** for this problem, because the CAD library doesn’t allow sizes less than 120nm, we will use L=100nm for both NMOS, W=120nm for top NMOS, and W=1 μm for bottom NMOS.

Here is the schematic of the the saturated-enhanced inverter cell view:

Here is the VTC plot of this circuit, DC sweeping the signal `INPUT`

from 0.0V to V_{DD} (1.2V). The green curve is the `OUTPUT`

signal.

To find the noise margin, we must first find V_{IL}, and V_{HH} which corresponds to the positions on the green curve where the slope is -1. We can find the slope by taking the derivative of the graph (in blue).

Where the slope equals to -1 is where input voltage is 0.257V and 0.733V for low and high, respectively. Thus, **V _{IL} = 0.257V** and

Next, by inspecting from the graph, the output low and high voltage are: **V _{OL} = 0.044V** and

Finally, we have all the values to compute the noise margin: