Lab 3
The Common-Source Amplifier
Compare the three loads of the common-source stage, see why a high-gain stage has no output bias of its own, and measure its poles, its zero and what a cascode does to them.
Contents
| Lab 3 at a glance | |
|---|---|
| Goal | Compare the three loads of the common-source stage, see why a high-gain stage has no output bias of its own, and measure its poles, its zero and what a cascode does to them. |
| Reproduces | Chapter 7 (Sections 7.1–7.2, Figures 7.1 and 7.3) and Chapter 9 (Worked example 9.1, Sections 9.6–9.7, Figure 9.9) |
| Time | 3 hours |
| Files | labs/lab3: cs_loads_tb, cs_mirror_tb, cs_ac_tb, cs_pz_tb, cascode_r_tb, cascode_i_tb (.sch) per PDK |
Hand calculation
With \(M_1\) of Worked example 6.1 at \(g_m/I_D\) = 15: (a) the gain of the diode-loaded stage from Eq. (7.4), with the load at 5 V⁻¹; (b) the gain of the current-source-loaded stage from Eq. (7.5), with \(g_{ds1}\) = 20.6 µS and \(g_{ds2}\) = 3.0 µS (SG13G2) or 14.7 µS and 1.15 µS (SKY130); (c) the dominant pole of the stage driven from 20 kΩ, \(f_{p1} \approx 1/(2\pi b_1)\), with \(b_1\) from Worked example 9.1.
Steps
- Three loads. Run
../../sim.sh cs_loads_tb.sch. The three stages of Figure 7.1(a)–(c) share one input sweep; the testbench prints the gain of each at its bias point and the output range over which the current-source load keeps half its gain. In the GUI,plot ga vs v(out_r) gb vs v(out_d) gc vs v(out_c)draws Figure 7.3. - A real current source.
cs_mirror_tb.schreplaces the ideal \(V_{BP}\) by a diode-connected PMOS carrying 41.9 µA, at the input voltage that put the ideal stage at \(V_{DD}/2\). Run it and note where the output goes, and how far it moves for 1 % more reference current and for 1 mV more input. - Frequency response. Run
../../sim.sh cs_ac_tb.sch: the stage of Worked example 9.1, driven from 20 kΩ into 2 pF, with the DC servo. It prints the DC gain, the −3 dB and unity-gain frequencies and the phase lag at \(f_u\), and the gate voltage the servo found. - Poles and zeros.
cs_pz_tb.schis the same circuit with the servo replaced by that gate voltage, becausepzcannot handle the servo’s 1 F capacitor. Run it;print alllists every pole and zero in rad/s as real,imaginary pairs. Divide by 2π and discard the pairs that coincide and the poles far above \(f_T\) (the pitfall of Section 9.7). - Cascode.
cascode_r_tb.schputs the CS stage and a cascode with the same devices side by side with a resistive load, \(R_S\) = 50 kΩ and 50 fF;cascode_i_tb.schdoes the same with ideal 41.9 µA loads, 20 kΩ and 2 pF. Each stage has its own servo, so both carry exactly 41.9 µA. Run both and compare the bandwidths and gains.
Checkpoint
Table 18.4. Lab 3 results (tt, 27 °C).
| Quantity | SG13G2 | SKY130 |
|---|---|---|
| Gain with resistor / diode / current-source load | 7.00 / 2.65 / 26.8 | 10.16 / 2.65 / 39.3 |
| Output range with at least half the gain (current-source load) | 0.15–1.02 V | 0.28–1.58 V |
cs_mirror_tb: output with the mirror instead of \(V_{BP}\) |
0.642 V (ideal: 0.600 V) | 0.897 V (ideal: 0.900 V) |
| Output change for 1 % more \(I_{REF}\) / 1 mV more input | +18.0 / −27.2 mV | +26.3 / −39.0 mV |
| \(R_S\) = 20 kΩ, 2 pF: DC gain, \(f_{-3\mathrm{dB}}\), \(f_u\), phase lag at \(f_u\) | 29.97, 1.546 MHz, 42.2 MHz, 113° | 42.39, 1.158 MHz, 49.1 MHz, 91° |
pz: poles; right-half-plane zero |
1.546 and 93.6 MHz; 3.05 GHz | 1.158 MHz and 1.245 GHz; 41.4 GHz |
| Resistive load, 50 kΩ, 50 fF: \(f_{-3\mathrm{dB}}\) CS → cascode | 18.5 → 28.3 MHz | 99.2 → 125.7 MHz |
| Ideal load, 20 kΩ, 2 pF: DC gain, \(f_u\), CS → cascode | 30.0 → 512, 42.2 → 41.5 MHz | 42.4 → 1019, 49.1 → 47.6 MHz |
Questions
- The diode-loaded gain is 2.65 in both PDKs, the current-source-loaded gain differs by 50 %. Why does the process cancel out of one and not the other?
- The mirror moves the SG13G2 output by 42 mV but the SKY130 output by only 3 mV. The mirror’s reference and output devices have the same size and current; what differs between them, and why does it matter more in SG13G2 (think of Section 8.2)?
- With the ideal load the cascode raises the gain 17 to 24 times but does not change \(f_u\). With the resistive load it raises the bandwidth instead. Explain both with the time constants of Section 9.4.
- Why does
pzreport a right-half-plane zero at \(g_m/(2\pi C_{dg})\) and not at \(g_m/(2\pi C_{gd})\), and which one did your simulation give?
Extension
In cs_mirror_tb.sch, double the length of both PMOS devices (keep \(W/L\)) and repeat step 2: does the 1 % sensitivity change, and does the systematic shift of the output? Then rerun cs_ac_tb.sch with \(R_S\) = 1 kΩ and 200 kΩ and compare the poles with the dominant-pole estimate of Figure 9.5.