Overview

GDS workflow DRC workflow LVS workflow DOCS workflow

Who

We are group 3 in the 2026 Advanced Integrated Circuits course. Test.

Why

To get an understanding of the design of advanced integrated circuits in CMOS technology, and to get an overview of the circuits needed to make a System-On-Chip.

How

The course consists, among other things, of a project divided into 5 (6 if tapeout) milestones. The idea is to design a temperature sensor. In the README below, each milestone will have a short description.

What

What Cell/Name
Schematic design/LELO_GR03_SKY130A/LELO_GR03.sch
Layout design/LELO_GR03_SKY130A/LELO_GR03.mag

Signal interface

Signal Direction Domain Description
VDD_1V8 Input VDD_1V8 Main supply
OSC_TEMP_1V8 Output VDD_1V8 Temperature dependent oscillation frequency
PWRUP_1V8 Input VDD_1V8 Power up the circuit
VSS Input Ground  

Key parameters

| Parameter | Min | Typ | Max | Unit | | :— | :—: | :—: | :—: | :—: | | Technology | | Skywater 130 nm | | | | AVDD | 1.7 | 1.8 | 1.9 | V | | Temperature | -40 | 27 | 125 | C | | Tc (conversion) | | ~30.5 | | µs | | Ts (sample rate) | | 100 | | ms | | Ileak (power-down) | | ~0.1 | 1 | nA | | Iact (active) | | ~79 | 100 | µA | | Iavg (average) | | ~24.1 | 50 | nA | | Kerr (1-pt cal) | | $\pm$22.0 | $\pm$10 | C | | Kerr (2-pt cal) | | $\pm$10.2 | $\pm$5 | C |

Milestone 1: The Bandgap

The bandgap OTA is a two-stage Miller OTA (see BANDGAP_OTA). The input NMOS transistors are low-threshold-voltage transistors, since they operate with the diode drop in the input common-mode voltage, reducing it from ~0.8V to ~0.5V over the temperature range of -40$^\circ$ to 125$^\circ$.

The bandgap circuit (see BANDGAP_CIRCUIT) uses a 1:8 BJT ratio (Q1 = 1$\times$, Q2 = 8$\times$). If we compare the voltages across the lower diode-connected BJTs, Q1 and Q2, the voltage difference will be proportional to the size difference and temperature:

\[V_{D1} - V_{D2} = \Delta V_{BE} = V_T \text{ln}\left (\frac{I_D}{I_{S1}}\right ) - V_T \text{ln}\left (\frac{I_D}{I_{S2}}\right ) = V_T \text{ln}(N)\]

Here $V_T = \frac{kT}{q}$. The V_CTAT voltage across Q1 will have a negative temperature coefficient and will be approximately linear over the temperature range of interest (-40$^\circ$ to 125$^\circ$). I_PTAT will be the current set by the voltage difference $\Delta V_{BE}$ over the resistor, which we denote as R1. The operational amplifier forces its inputs to be equal, resulting in a voltage drop $V_{D1} - V_{D2} = \Delta V_{BE}$ across R1. I_PTAT will thus be $\Delta V_{BE} / R1$.

The plots below show the corner simulations over temperature for PTAT current, CTAT voltage, power-down leakage current, and active supply current. The PTAT current varies significantly across process corners, as verified by the resistance variation of resistor R1 across the different corners. We tested this by replacing R1 with a generic resistor with no process variations and observed much less variation, with a worst-case error of less than 10% compared to 30%. The power-down leakage stays well below 1 nA across all corners.

[Image: SVG not converted]

Milestone 2: The Oscillator

The PTAT current from the bandgap charges a timing capacitor (12 MIM caps, see OSCILLATOR), and the voltage across the capacitors feeds into the COMPARATOR alongside the reference voltage V_CTAT. Then, if the capacitor voltage exceeds V_CTAT, the comparator fires, and a reset pulse discharges the capacitor through an NMOS switch, restarting the cycle. The comparator output feeds into a D flip-flop, which divides the frequency by 2 to produce a clean square wave. Since both the charging current (PTAT) and the threshold voltage (CTAT) are proportional to temperature, the resulting output frequency is approximately linear in temperature. Using the equation for current through a capacitor:

\[i = C \frac{dV}{dt} \Rightarrow dt = C \frac{dV}{i} \Rightarrow f(T) = \frac{1}{dt} = \frac{I_{PTAT}(T)}{C \cdot V_{CTAT}(T)}\]

Temperature error estimation

Since there is no FSM and counter yet, we estimate the temperature measurement error directly from the simulated frequency. The idea is to invert the frequency-to-temperature relationship: given a measured frequency, we compute what temperature a calibrated system would report, and compare that to the actual simulation temperature.

1-point calibration uses a single calibration point at 25$^\circ$C. We take the nominal frequency-vs-temperature slope from the typical corner (KttVt) and apply it to all corners. For each corner, we compute the offset at 25$^\circ$C and convert frequency back to temperature using:

\[T_{meas} = \text{slope}_{nom} \cdot f + \text{offset}_{25}\]

The calibration assumes all chips share the same slope, which breaks down at extreme PVT corners.

2-point calibration fits a line through two calibration points at 25$^\circ$C and 85$^\circ$C for each individual corner. This per-chip calibration removes slope variation and only leaves residual non-linearity as error:

\[T_{meas} = \frac{85 - 25}{f_{85} - f_{25}} \cdot (f - f_{25}) + 25\]

The plots below show the oscillator performance across PVT corners. All corners pass the 2-point calibration spec of $\pm$5$^\circ$C with margin (max 2.1$^\circ$C error). The 1-point calibration meets $\pm$10$^\circ$C for most corners, with the extreme corners KffVh and KssVl slightly exceeding the limit.

[Image: SVG not converted]

Milestone 3: Counter

The goal is to measure the frequency of the oscillator. We assume access to an accurate 32768 Hz clock source. The approach is to power up the oscillator for a fixed number of reference clock cycles and count the output pulses. For example, counting 128 pulses over 2 periods of the 32768 Hz clock gives a frequency of approximately 2.09 MHz. Once we have the frequency, we can calculate the temperature.

The digital block (temp_sens.sv) implements a 4-state FSM with a dual-edge counter that captures both rising and falling edges of the oscillator output for 2$\times$ resolution. The FSM powers up the oscillator for one reference clock period (~30 µs), lets the counter settle (CDC safety), then latches the count. See the FSM state diagram below.

The behavioral simulation fits a 2nd-order polynomial to the SPICE-characterized oscillator frequency, then sweeps temperature from -40$^\circ$C to 125$^\circ$C. The plots below show the measured count and calibration error across all PVT corners:

[Image: SVG not converted]

Implementation

BANDGAP_OTA

[Image: SVG not converted]

BANDGAP_CIRCUIT

[Image: SVG not converted]

COMPARATOR

[Image: SVG not converted]

OSCILLATOR

[Image: SVG not converted]

TempSens FSM

stateDiagram-v2
    [*] --> IDLE
    IDLE --> COUNT : i_start
    COUNT --> WAIT
    WAIT --> CAPTURE
    CAPTURE --> IDLE

| State | o_pwrup_osc | Description | | :— | :—: | :— | | IDLE | 0 | Waiting for trigger. Counter held in reset. | | COUNT | 1 | Oscillator on. Dual-edge counter runs for one 32768 Hz period. | | WAIT | 0 | Oscillator off. Counter values settle (CDC safety). | | CAPTURE | 0 | Count latched into output register. |

Install

Clone LELO_GR03_SKY130A

To install, do the following

python3 -m pip install cicconf
git clone --recursive https://github.com/analogicus/lelo_gr03_sky130a lelo_gr03_sky130a
cicconf --rundir ./ --config lelo_gr03_sky130a/config.yaml clone --https

Schematics

LELO_GR03_SKY130A

BANDGAP_OTA

[Image: SVG not converted]

BANDGAP_CIRCUIT

[Image: SVG not converted]

COMPARATOR

[Image: SVG not converted]

OSCILLATOR

[Image: SVG not converted]

Simulations

  • TOC

LELO_GR03_SKY130A

BANDGAP_OTA

BANDGAP_CIRCUIT

COMPARATOR

OSCILLATOR

Layout

GDS

GLTF

3D Model