Power Monitor ZCU102

The power_monitor tool provides a Python interface to accurately read INA226 power rails via the hwmon subsystems on the Xilinx Zynq UltraScale+ ZCU102 development board.

The design for this tool leverages the approach in the Xilinx EDA365 article: Accurate Design Power Measurement Made Easier by Don Matson and Luis Bielich.

---

Power Calculation Formulas

The total power of the board is divided into three major domains: Processing System (PS), Programmable Logic (PL), and Multi-Gigabit Transceivers (MGT). The formulas used to calculate total power are derived from the reference article:

• PS Power = VCCPSINTFP + VCCPSINTLP + VCCPSAUX + VCCPSPLL + VCCPSDDR + VCCOPS + VCCOPS3 + VCCPSDDRPL
• PL Power = VCCINT + VCCBRAM + VCCAUX + VCC1V2 + VCC3V3
• MGT Power = MGTRAVCC + MGTRAVTT + MGTAVCC + MGTAVTT
• Total Power = PS Power + PL Power + MGT Power

Note: While some legacy implementations vary where VCC3V3 is categorized, this tool adds it to PL as indicated by the formula structure in the article text.

---

INA226 Power Rails for ZCU102

The ZCU102 has shunt resistors placed on specific power rails, monitored by TI INA226 chips. These are accessible via the I2C bus exported to /sys/class/hwmon on PetaLinux systems.

Here is the table of the relevant rails (referenced from the article and board parameters):

Sensor Name	Rail / Domain Name	Category
ina226_u76	VCCPSINTFP	PS
ina226_u77	VCCPSINTLP	PS
ina226_u78	VCCPSAUX	PS
ina226_u87	VCCPSPLL	PS
ina226_u93	VCCPSDDR	PS
ina226_u88	VCCOPS	PS
ina226_u15	VCCOPS3	PS
ina226_u92	VCCPSDDRPL	PS
ina226_u79	VCCINT	PL
ina226_u81	VCCBRAM	PL
ina226_u80	VCCAUX	PL
ina226_u84	VCC1V2	PL
ina226_u16	VCC3V3	PL
ina226_u65	VADJ_FMC	PL
ina226_u85	MGTRAVCC	MGT
ina226_u86	MGTRAVTT	MGT
ina226_u74	MGTAVCC	MGT
ina226_u75	MGTAVTT	MGT

---

Usage Example

Once the package is installed on the ZCU102 PetaLinux environment, you can use it in your Python scripts.

1. Simple Single Read

from power_monitor import PowerMonitor

# Initialize the monitor
pm = PowerMonitor()

# Read the aggregated power domains once
ps, pl, mgt, total = pm._read_once(verbose=True)

print(f"Total Power: {total:.3f} W")

2. Background Measurement (Non-Blocking)

Useful for measuring power while a separate workload runs.

import time
from power_monitor import PowerMonitor

pm = PowerMonitor(interval=0.5) # Sample every 0.5 seconds

print("Starting background power measurement...")
pm.start()

# Do your heavy workload here (e.g., neural network inference)
time.sleep(3.0) 

# Stop monitoring and retrieve records
pm.stop()

# Summarize and save
avg_ps, avg_pl, avg_mgt, avg_total = pm.average()
print(f"Average Total Power during workload: {avg_total:.3f} W")

pm.save_samples("power_log.csv")

---

Installation

From source (PetaLinux 2022.2)

git clone <repository-url>
cd power_monitor
pip install .

---

Running Automated Tests

The tests are specifically written to target the hwmon structures of the ZCU102 running PetaLinux.

Requirements:

• A physical ZCU102 board running the PetaLinux 2022.2 image.
• pytest installed.

To run the tests:

1. Secure Shell (SSH) into the ZCU102 or use the serial terminal.
2. Navigate to the power_monitor project root.
3. Install pytest if needed: pip install pytest
4. Run the suite:

   python -m pytest tests/