940 LELO-GR03 :: Quicker, easier and cheaper to make your own chip!

Repository: https://github.com/analogicus/lelo_gr03_sky130a
Documentation: https://analogicus.github.io/lelo_gr03_sky130a

Who

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

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.

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)		~0.2	1	nA
Iact (active)		~51	100	µA
Iavg (average)		~15.8	50	nA
Kerr (1-pt cal)		±8.9	±10	C
Kerr (2-pt cal)		±8.9	±5	C

Milestone 1: The Bandgap

The bandgap OTA is a two-stage Miller 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° to 125°.

The bandgap circuit uses a 1:8 BJT ratio (Q1 = 1×, Q2 = 8×). 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° to 125°). 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.

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°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°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°C and 85°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 ±5°C with margin (max 2.1°C error). The 1-point calibration meets ±10°C for most corners, with the extreme corners KffVh and KssVl slightly exceeding the limit.

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× resolution. The FSM powers up the oscillator for one reference clock period (~30 µs), lets the counter settle (CDC safety), then latches the count.

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

Milestone 4: The physical design

In milestone 4 we implemented the physical design of our oscillator. The layout of the oscillator, LELO_GR03.mag passes the drc check, but there is some port trouble in the LVS check. It can't seem to separate VDD_1V8 and VSS causing an issue. The submodules, BANDGAP_CIRCUIT.mag, BIAS_CIRCUIT.mag and COMPARATOR.mag passes both drc and lvs check. The issue arises when we connect everything together in LELO_GR03. The layout can be seen in the figure below. The Left part of the image comprises the bandgap circuit, the lower right part is the comparator and D-flip-flop for the relaxation oscillator and the middle area is the bias circuit generating bias voltages for the OTA (in the bandgap circuit) and the comparator (lower rigth side). The layout is made to fit in a Tiny Tapeout template with the digital circuitry being supposed to fit in the upper right corner.

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.

#	Input	Output
0	PWRUP	OSC_TEMP
1
2
3
4
5
6
7

940 LELO-GR03

940 : LELO-GR03

Who

Why

How