Simple Current Mirrors I out I in v1 r out Q2 Q1 A simple CMOS current-mirror. Basic Current Mirrors and Gain Stages Ken Martin • Q 1 is a ‘diode-connected transistor; its small-signal model is simply a resistor of Dept. of Elec. and Comp. Eng. University of Toronto Toronto, Canada M5S 1A4 • At low-frequencies, MOS transistors are unidirectional; voltage changes at drain or source can not effect gate current or voltage. Thus, when calculating output impedance one may assume V gs2 = 0 . size 1 ⁄ g m . Q1 martin@eecg.toronto.edu (416) 978-6695 1 ---------g m1 0V Q2 ix r ds2 v gs2 g v m2 gs2 vx (a) ix r ds2 vx (b) (a) A small-signal model for finding the output impedance of the simple currentmirror of and (b) a simplified small-signal model. © D.A. Johns/K.W. Martin, 1995 1 © D.A. Johns/K.W. Martin, 1995 2 Source Follower Common-Source Amplifier with Active Load Active Load Q3 Q2 V in V out I bias I r out Q1 V in Q V out bias Q 1 Q2 3 Active Load A common-source amplifier with a current-mirror active load. A source-follower stage with a current-mirror used to supply the bias current. KL i • r out = r ds1 || r ds2 where r dsi = --------I Di • This is one of the few circuits where the body-effect has a major effect on the gain. vd1 Rin vout vin = vg1 rds1 vgs1 v gs1+ - vin g m1 v gs1 gs1vs1 g m1v gs1 R 2 = r ds1 || r ds2 vs1 rds2 A small-signal equivalent circuit for the common-source amplifier. • A V = – g m1 r out = – g m1 ( r ds1 || r ds2 ) © D.A. Johns/K.W. Martin, 1995 3 where g m1 = 2µ n C ox ( W ⁄ L ) 1 I D1 The low-frequency model of the source-follower amplifier. © D.A. Johns/K.W. Martin, 1995 4 vout = vs1 Common-Gate Amplifier • The current source g s1 v s1 is equivalent to a resistor of size 1 ⁄ g s1 . This allows us to simplify the model leading to Active Load Q3 vin Q2 V out r out I bias vgs1 g m1v gs1 Q1 r in V bias vout R s1 = r ds1 ||r ds2 1|| ⁄ g s1 V in A common-gate amplifier with a current-mirror active load. An equivalent small-signal model for the source-follower. • This stage is used extensively for high-speed design g m1 A V = ---------------------------------------------------------g m1 + g s1 + g ds1 + g ds2 vout (1) vgs1 γ gm where g s1 = -------------------------------------- is approximately g m1 ⁄ 5 to g m1 ⁄ 10 2 V SB + 2φ F gm1vgs1 vs1 rds1 gs1vs1 RL rin Rs vin The small-signal model of the common-gate amplifier at low frequencies. • Note that v s1 = – v gs1 and that therefore the two current sources can be combined. This simplification can always be done at low frequencies when considering signals at the drain or source irrespective of what is connected to the gate. This often allows one to do analysis ignoring the body-effect and then taking it into account at end of analysis. © D.A. Johns/K.W. Martin, 1995 5 © D.A. Johns/K.W. Martin, 1995 6 vout vgs1 rds1 (gm1+gs1)vs1 vs1 Source-Degenerated Current Sources RL rin I out I in Rs vin A simplified small-signal model of the common-gate amplifier. • First analyzing for admittance looking into the source, we have g m1 + g s1 + g ds1 g m1 ′ g in = ----------------------------------------- ≅ -------------------------------1 + g ds1 ⁄ G B 1 + g ds1 ⁄ G B V1 r out Q1 Q2 Rs Rs A current-mirror with source degeneration. (2) or equivalently RL 1 r in ≅ ----------- 1 + --------- g m1 ′ r ds1 1 ---------g m1 where R L = 1 ⁄ G L is the output impedance of the active load. • Note that when R L is large, as is the case for ‘better’ active loads, r in can be considerably greater than 1 ⁄ g m1 • Using the admittance divider rule, we now have v s1 GS ------- = ---------------------v in G S + g in Gs g m1 ′ ≅ ---------------------------------------------------- -------------------------G L + g ds1 g m1 ′ G s + -------------------------------------1 + ( g ds1 ⁄ G L ) © D.A. Johns/K.W. Martin, 1995 7 v gs g m2 ′v gs r ds2 vx vs Rs ix Rs The small-signal model for the source-degenerated current source. (4) • Solving for r out , we get and g m1 + g s1 + g ds1 v s1 v out Gs A V = ------- --------- = ---------------------------------------------------------- -------------------------------------------G L + g ds1 v in v s1 g m1 + g s1 + g ds1 G s + -------------------------------------------1 + ( g ds1 ⁄ G L ) ix 0V (3) vx r out = ----- = r ds2 [ 1 + R s ( g m2 ′ + g ds2 ) ] ≅ r ds2 ( 1 + R s g m2 ′ ) ix (6) • This formula is often very useful in estimating output impedances quickly. (5) • For a bipolar transistor this relationship is modified to be r out = r ds2 [ 1 + g m2 ( R s || r π ) ] r out ≤ r ds2 [ 1 + g m2 r π ] ≅ r ds2 [ 1 + β ] © D.A. Johns/K.W. Martin, 1995 8 (7) (8) Cascode Current Mirrors Wilson Current Mirror V out I in I out r out 2 ( V eff + V tn ) Q3 ( V eff + V tn ) Q4 I in r in Q2 Q3 Q4 Q1 Q2 ( V eff + V tn ) Q1 A cascode current-mirror. The Wilson current-mirror. • Using analysis of source-degenerated mirror, we have r out = r ds4 [ 1 + r ds2 ( g m4 ′ + g ds4 ) ] ≅ r ds4 ( 1 + r ds2 g m4 ′ ) (9) ≅ r ds4 ( r ds2 g m4 ′ ) • Assuming all transistors have the same current densities and therefore effective gate-source voltages, V eff , we have V GSi = V eff + V tn and therefore V G1 = V eff + V tn V G3 = V G4 = 2 ( V eff + V tn ) (10) V S4 = V G4 – V GS4 = V eff + V tn and since to keep Q 4 in active region, we need V DS4 > V eff , we need V out > V tn + 2V eff ∼ 1.3V I out (11) • This is an example of series-series feedback which increases output impedance by 1 + A L where A L is the loop gain. For Wilson current-mirror, the loop-gain is given by g m1 ( r ds1 || r in ) (12) A L ≅ -------------------------------------2 where r in is the output impedance of the input current source, I in . Therefore g m1 r g m1 ( r ds1 || r in ) ds1 r out ≅ r ds4 -------------------------------------- ∼ r ds4 --------------------- 2 2 • This current mirror also requires V tn + 2V eff across to keep the cascode transistor out of the triode region. • This is for n-channel mirror; p-channel mirror would also require about 1.3V leaving 0.4V for signal swings assuming 3V supply voltages. This is not acceptable. © D.A. Johns/K.W. Martin, 1995 9 (13) © D.A. Johns/K.W. Martin, 1995 10 Cascode Gain Stage MOS Differential Pair Ibias Ibias1 Vout Vbias Q2 Vin Q1 CL I D1 Q2 Vin Q1 V Vbias (a) + V Q1 Vout Ibias2 I D2 Q2 I bias CL (b) - A MOS differential pair. (a) A telescopic cascode amplifier and (b) a folded-cascode amplifier. • Currently very popular because a) gain of single stage is increased without significantly decreasing speed, and b) minimizes short-channel effects of modern technology. i d1 = i s1 + v • Telescopic implementation is fastest (assuming n-channels used), folded-cascode has compatible bias voltages at input and output. r s1 • The low-frequency gain is on the same order of the gain of a cascode of two gain stages having simple active loads (assuming the active load of the cascode stage is a ‘good’ current source. 1 2 2 (14) A V ∼ --- g m r ds 2 • Note, that at low frequencies there is considerable gain to v s2 , on the order of g m r ds ⁄ 2 . At higher frequencies this gain decreases to -1. © D.A. Johns/K.W. Martin, 1995 11 i d2 = i s2 v i s1 i s2 + - r s2 - v –v --------------------r s1 + r s2 The small-signal model of a MOS differential pair. v in g m1 v in i d1 = i s1 = – i d2 = --------------------- = ----------------------------------------- ≅ ---------- v in 1 ⁄ g m1 + 1 ⁄ g m2 2 r s1 + r s2 © D.A. Johns/K.W. Martin, 1995 12 (15) Bipolar Current Mirrors MOS Differential-Pair with Active Load Q3 Q4 i s1 r out i d4 i s1 v in Q1 i s1 v out I in I out Q1 Q2 A simple bipolar current-mirror. Q2 1 I out = ------------------------ I in ≈ ( 1 – 2 ⁄ β )I in (1 + 2 ⁄ β) I bias I out1 I in A differential-input single-ended-output MOS gain stage. r out = r ds4 || r ds2 (16) A V = g m1 r out = g m1 ( r ds4 || r ds2 ) (17) (18) I out2 Q3 Q1 Q4 Q2 A current-mirror with less inaccuracies caused by finite base currents. • Note that all nodes are low impedance (i.e. impedances on the order of 1 ⁄ g m ) except the output node. 2 I out ≈ I in ( 1 – 2 ⁄ β ) I in (19) I out Q3 Q1 Q2 Re Re A current-mirror with emitter degeneration. © D.A. Johns/K.W. Martin, 1995 13 © D.A. Johns/K.W. Martin, 1995 14 V R ≈ 0.2V e r out = r o2 ( 1 + g m2 ( R e || r π ) ) ≅ r o2 ( 1 + g m2 R e ) (20) V Re V Re R e = ----------- ≅ ----------I e2 I c2 (21) Bipolar Differential Pair I C1 Using the relationship g m2 = I c2 ⁄ V T , where V T = kT ⁄ q ≅ 26 mV at 300°K, gives V Re r out ≅ r o2 ( 1 + g m2 R ) = r o2 1 + ----------- ≅ 11r o2 (22) e VT • More importantly, the emitter degeneration greatly minimizes the thermal noise due to the base resistance which is often the dominant noise source in wide-band bipolar circuits V Iout Q1 Q2 V - + -V be2 I EE A bipolar differential pair. ( V BE ⁄ V T ) IC = IS e V id = V BE1 – V BE2 Iin Iout Q3 Q4 Q3 Q4 Q1 Q2 Q1 Q2 I C1 -------- = e I C2 V + – V - ------------------ VT = e (23) (24) V id -------- VT (25) αI EE = I C1 + I C2 (26) αI EE I C1 = -----------------------------------( – V id ⁄ V T ) 1+e αI EE I C2 = ---------------------------------( V id ⁄ V T ) 1+e (b) (a) + V be1 High-Impedance Bipolar Current Mirrors Iin + I C2 High output impedance current-mirrors (a) Cascode (b) Wilson. • For both mirrors r out ≅ ( βr o ) ⁄ 2 (27) (28) I C1, I C2 • For cascode mirror, I out ≅ 1 – 4 ⁄ β ; for Wilson current mirror, I out ≅ 1 – 2 ⁄ β 2 . Moral: always use Wilson current mirrors for bipolar design αI EE I C2 I C1 0.5αI EE –4 –2 0 2 Collector currents for a bipolar differential-pair. © D.A. Johns/K.W. Martin, 1995 15 © D.A. Johns/K.W. Martin, 1995 16 4 V id ⁄ V T Frequency Response sC gs1 + g m1 A ( s ) = A ( 0 ) ----------------------------------s s2 1 + ----------- + ------ω0 Q ω 2 0 Common-Source Stage Rin v1 Cgd1 where vout + C gs1 g v v gs1 m1 gs1 - vin R2 ω0 = C2 1 ω – 3dB ≅ ----------------------------------------------------------------------------------------------------------------------------R in [ C gs1 + C gd1 ( 1 + g m1 R 2 ) ] + R 2 ( C gd1 + C 2 ) (29) 1 ω – 3dB ≅ ----------------------------------------------------------------R in [ C gs1 + C gd1 ( 1 + A ) ] (30) Rin C' in (32) (33) and π – ----------------------2 4Q – 1 (34) % overshoot = 100e • Source follower is fast, but can have overshoot especially when R in is small and C L is large. Common-Gate Stage Source Follower iin G in ( g m1 + G s1 ) -------------------------------------------------------------------C gs1 C s + C' in ( C gs1 + C s ) G in ( g m1 + G s1 ) [ C gs1 C s + C' in ( C gs1 + C s ) ] Q = -----------------------------------------------------------------------------------------------------------------------G in C s + C' in ( g m1 + G s1 ) + C gs1 G s1 A small-signal model for high-frequency analysis of the common-source amplifier. vg1 (31) • No Miller effect so time constant at input node seldom dominates. This makes it typically much faster than common source stage. Time constant at output (i.e. ( r ds ⁄ 2 ) ( C L + C db ) ) usually dominates. Yg Cgs1 vgs1 g m1v gs1 vout • Time constant at input (at high frequencies) is on the order of ( 1 ⁄ g m )C gs . This can be important when determining approximate second pole frequency. Cascode Gain Stage Rs1 C' in = C in + C gd1 Cs R s1 = r ds1 ||r ds2 1|| ⁄ g s1 • No Miller effect so often faster than common-source stage. 2 )(C • Time constant at output (i.e ( g m r ds db2 + C L ) ) dominates. A simplified equivalent small-signal model for the source-follower. • Time constant at source of cascode transistor is approximately L 22 τ s2 ∼ ------------------µ p V eff2 (35) This fundamental relationship assumes a folded-cascode stage with Q 2 being pchannel © D.A. Johns/K.W. Martin, 1995 17 © D.A. Johns/K.W. Martin, 1995 18 References • P. R. Gray and R. G. Meyer, Analysis and Design of Analog Integrated Circuits, 3rd ed., John Wiley and Sons, New York, 1993. • D. Johns and K. Martin, Analog Integrated Circuits, Wiley, 1997. • A.S. Sedra and K.C. Smith, Microelectronic Circuits, 3rd ed., Holt, Rinehart and Winston, New York, 1991. © D.A. Johns/K.W. Martin, 1995 19