A long-channel EKV MOSFET pair drives a two-stage Miller op-amp, which in turn becomes a
+ self-running relaxation oscillator. Every curve below is real simulation output from the
+ compiled Dyad models — no idealized equations, no fitted overlays.
+
+
+
–
Op-amp open-loop gain
+
–
Output swing (5 V rail)
+
–
Oscillator frequency
+
EKV
Single-expression FET model
+
+
+
+
+
+
+
+
+
01
The MOSFET models
+
+ NMOS and PMOS use the charge-based EKV formulation: one smooth
+ expression spans weak, moderate and strong inversion — no operating-region branches. The
+ output family shows clean saturation; the transfer curves show the continuous
+ sub-threshold-to-strong transition, with the p-channel scaled by its lower transconductance.
+
+
+
+
NMOS output characteristics
+
Drain current vs VDS, stepped gate bias — the classic saturation family
+
+
+
+
Transfer characteristic
+
+
ID vs gate drive at VDS = 1 V — NMOS vs PMOS
+
+
+
+
+ NMOS KP = 200 µA/V²
+ PMOS KP = 70 µA/V²
+ VT0 = 0.4 V
+ L = 1 µm
+ Wright-ω inversion of the EKV charge equation keeps the model finite across the whole
+ floating-point range.
+
+
+
+
+
+
02
Two-stage CMOS op-amp
+
+ An NMOS differential pair with a PMOS mirror load feeds a common-source second stage, Miller
+ compensated. Sized for near-zero systematic offset, it reaches ~90 dB of open-loop gain
+ and swings nearly rail to rail. Closed as a unity-gain follower it tracks its input to a
+ fraction of a millivolt.
+
+
+
+
Open-loop transfer (VTC)
+
Output vs differential input — the steep crossover is the DC gain
+
+
+
+
Unity-gain follower tracking
+
Output follows a 50 kHz input about mid-rail
+
+
+
+
+
+
+
+
03
Relaxation oscillator
+
+ With positive feedback the op-amp becomes a Schmitt-trigger comparator; an RC on the
+ inverting input charges and discharges between the trip points, so the loop has no stable
+ DC state and free-runs. The square-wave output and the exponential capacitor ramp are shown
+ together — the ramp turns around exactly at the Schmitt thresholds.
+
+ f ≈ 9.4 kHz
+ T ≈ 106 µs
+ β = R₂/(R₁+R₂) ≈ 0.25
+ Measured period sits within ~5% of the ideal 2·R·C·ln((1+β)/(1−β)), the gap being the
+ finite comparator switching speed.
+
The exact topology of the simulated CMOSOpAmp Dyad component — eight EKV FETs, one resistor, two capacitors, self-biased from the rails
+
+ NMOS
+ PMOS
+ node A (stage-1 output)
+
+
+
+ M1/M2 form the differential pair on M5's tail current, loaded by the M3/M4 mirror; stage-1
+ output is node A. M6 (sized 2 W for zero systematic offset) drives the output against
+ the M7 sink, Miller-compensated by Cc. Rb through diode-connected M8
+ sets the reference that M5 and M7 mirror — the amplifier biases itself from the rails.
+ Bulk connections (NMOS→VSS, PMOS→VDD) are omitted for clarity.
+
+
Open-loop transfer (VTC)
Output vs differential input — the steep crossover is the DC gain