Design and Implementation of Multi-Channel Temperature Measurement System of Thermal Test Chip Based on Diode Temperature-Sensitive Arrays

Abstract

When chips perform numerous computational tasks or process complex instructions, they generate substantial heat, potentially affecting their long-term reliability and performance. Thus, accurate and effective temperature measurement and management are crucial to ensuring chip performance and lifespan. This paper presents a multi-channel temperature measurement system based on a diode temperature-sensitive array thermal test chip (TTC). The thermal test chip accurately emulates the heat power and thermal distribution of the target chip, providing signal output through row and column address selection. The multi-channel temperature measurement system centers around a microcontroller and includes voltage signal acquisition circuits and host computer software. It enables temperature acquisition, storage, and real-time monitoring of 16 channels in a 4 × 4 array thermal test chip. During experiments, the system uses a constant current source to drive temperature-sensitive diodes, collects diode output voltage through multiplexers and high-precision amplification circuits, and converts analog signals to digital signals via a high-speed ADC. Data transmission occurs via the USB 2.0 protocol, with the host computer software handling data processing and real-time display. The test results indicate that the system accurately monitors chip temperature changes in both steady-state and transient thermal response tests, closely matching measurements from a semiconductor device analyzer, with an error of about 0.67%. Therefore, this multi-channel temperature measurement system demonstrates excellent accuracy and real-time monitoring capability, providing an effective solution for the thermal design and evaluation of high power density integrated circuits.

1. Introduction

In the fields of artificial intelligence and supercomputing, chips frequently handle high-intensity computational tasks, leading to significant heat generation. As operating frequencies rise, thermal output increases, adversely affecting long-term reliability, performance, and integration [1,2,3]. Traditional temperature testing methods, such as thermocouple sensors placed on or near the chip surface, measure temperature by detecting voltage changes but can interfere with chip operation. Infrared thermal imaging offers a non-contact solution by using infrared cameras to visualize temperature distribution, though it is limited by resolution and accuracy. Another common approach is thermistor-based temperature measurement, which gauges temperature by monitoring resistance changes. While this method is accurate within a specific range, it requires complex calibration. Effective thermal management is crucial for maintaining the optimal performance of AI and supercomputing chips. This study references established thermal testing standards such as JEDEC JESD51-1 through JESD51-12 to ensure compliance and provide a baseline for evaluating the system’s performance. Compared to these standards, the proposed system enhances precision and real-time monitoring capabilities, addressing limitations in conventional methodologies.

This study builds upon previous research on thermal test chips, including the pioneering TTC-1001 series developed by Thermal Engineering Associates, which laid the foundation for high heat flux thermal testing. However, these solutions face limitations in terms of channel scalability, real-time measurement, and system integration. To address these challenges, this study develops a novel thermal simulation chip integrating heating and temperature measurement elements. By configuring its heating units, it improves upon existing designs through enhanced multi-channel acquisition circuits (e.g., combining high-speed ADCs with precision amplifiers) and optimized data transmission protocols, offering unique advantages in thermal design and simulation. The thermal test chip can be packaged like the target chip, accurately simulating its power distribution post-packaging. It is becoming crucial for chip packaging and high power density integration thermal design and evaluation, and increasingly necessary for high-density integration thermal simulation [4,5].

Mainstream thermal test chip technology currently uses temperature-sensitive diode structures, converting temperature signals to voltage output for testing. For instance, Thermal Engineering Associates (TEA) developed the TTC-1001 series, which has become a benchmark for high heat flux testing. However, these systems face limitations in transient thermal response monitoring, scalability for large diode arrays, and integration with high-speed data acquisition systems. The RQS-TTC-SC01 chip offers improvements but remains constrained by its relatively low precision under dynamic conditions. Thermal Engineering Associates (TEA) in the USA introduced thermal test chips based on diode temperature measurement units, such as the TTC-1001, now mainstream products for evaluating packaging and high heat flux density integration [6]. In China, research and application in this area began later. Recent demand for high integration has increased research applications in this area, such as the RQS-TTC-SC01. Thermal test chips use an array structure of diodes for temperature measurement, achieving signal output via row and column address selection [7]. The known array structure in thermal test chips can reach up to 32 rows and 32 columns, totaling 1024 temperature sensor units. However, existing test circuits and instruments cannot meet actual thermal testing needs, requiring additional control circuit designs. The design parameters of different thermal test chips vary, and the operating temperature range is large, causing wide changes in electrical signals output by temperature measurement elements. The readout system must handle wide-range input voltage while ensuring temperature measurement accuracy over a large range [8]. Additionally, in thermal testing experiments related to heat dissipation and packaging, analyzing the process of chip thermal equilibrium establishment often requires high-speed readout capabilities to observe transient thermal changes. This poses significant challenges for the readout system.

This paper addresses the aforementioned issues by designing and fabricating a multi-channel temperature measurement system for a thermal test chip, leveraging the properties of a diode-based temperature-sensitive array. The system is centered around a microcontroller (MCU) and includes a voltage signal acquisition circuit for the diode array. The working sequence enables gating scans, signal amplification, and analog-to-digital conversion. The upper computer software, developed in Python, handles temperature acquisition, storage, and real-time monitoring of the 16 channels in the 4 × 4 thermal test chip array. The temperature monitoring system includes data processing and USB communication, enabling real-time transmission of processed temperature data to a PC [9]. By adjusting the parameters, temperature changes on the thermal test chip can be easily monitored, making the system adaptable to various thermal testing requirements. It provides a practical method for thermal evaluation and testing of electronic devices and the system [10]. The proposed system surpasses existing solutions by enabling high-resolution real-time monitoring with ±0.1 °C accuracy and a scalable architecture supporting 48 × 48 diode arrays. It addresses key limitations in transient thermal response and precision, providing a significant advancement for thermal management research [11].

2. System Design

To enable multi-channel temperature measurement for array structures, we designed a system for a thermal test chip using a microcontroller. The system consists of three main components: the acquisition circuit module, the microcontroller module, and the upper computer software module. The block diagram of the readout system structure of the thermal test chip is shown in Figure 1. The acquisition circuit scans and gates the diode temperature-sensitive array in the thermal test chip, applies a current source drive, reads the diode output voltage, amplifies the output signal, and performs analog-to-digital conversion. The microcontroller receives instructions from the upper computer, controls the acquisition circuit’s working sequence, and transmits the acquired digital signals to the upper computer via the USB interface. The upper computer software interface receives and sends user instructions, saves the received temperature data in files, and displays the processed data in real-time on the screen. Integration of adaptive amplification circuits for dynamic range adjustment, ensuring high accuracy across various thermal ranges. A high-speed ADC is implemented for millisecond-level response times. A Python-based software platform with graphical visualization is developed, enhancing usability and data analysis capabilities.

2.1. Acquisition Circuit Design

The acquisition circuit drives the current source, gates the diode temperature-sensitive array, and reads the output voltage. The voltage is amplified and sent to the analog-to-digital converter (ADC), where it is converted into digital signals for transmission and storage. The microcontroller reads these digital signals through I/O pins.

Figure 2 shows the overall structure of the acquisition circuit module. The circuit uses a single-channel multiplexing gating mode, employing the DG406 multiplexer chip as the analog switch. The diode units are gated by address signals, allowing precise control of the 48 × 48 array structure. A constant current source drives the temperature-sensitive diodes. Under constant current drive, the forward voltage drop V_D of the diode has a linear relationship with temperature. This circuit structure is simple, has good anti-interference capability, and provides a stable output voltage. The LM334 three-terminal adjustable current source chip serves as the constant current source. Changing the value of the external resistor adjusts the constant current source output within the range of 1 μA to 10 mA. The positive voltage of the diode varies widely due to the different design parameters of the thermal test chip diodes and the size of the constant current source. The high-precision voltage regulator LT3024 provides the positive voltage V_ref0 to match the diode output voltage with the amplifier input voltage range. The output voltage of the diode ${\mathrm{V}}_{\mathrm{i}\mathrm{n}}={\mathrm{V}}_{\mathrm{r}\mathrm{e}\mathrm{f}0}{-\mathrm{V}}_{\mathrm{D}}$. Adjusting the value of the matching resistor (R₂) changes the output value of the reference voltage V_ref0. The output voltage range is adjustable from 1.2 V to 6.5 V, with a voltage source noise of 20 μV_RMS.

The OPA2613 chip implements the amplification circuit. The input uses a voltage follower circuit structure. The voltage follower does not amplify voltage; it only extracts the voltage signal. The voltage follower has a large input impedance and a very small input current to avoid affecting the constant current source diversion, ensuring temperature measurement accuracy (when IM = 10 μA, the input current is 45 nA, ≦0.5%IM). The two-stage amplification circuit structure is identical, and the amplification factor depends only on the values of resistor R_s and adjustable resistor S_R1. ${\mathrm{V}}_{\mathrm{o}\mathrm{u}\mathrm{t}1}={{\displaystyle \frac{{\mathrm{R}}_{\mathrm{s}}+{\mathrm{S}\mathrm{R}}_1}{{\mathrm{R}}_{\mathrm{s}}}}\mathrm{V}}_{\mathrm{r}\mathrm{e}\mathrm{f}1}{-{\displaystyle \frac{{\mathrm{S}\mathrm{R}}_1}{{\mathrm{R}}_{\mathrm{s}}}}\mathrm{V}}_{\mathrm{i}\mathrm{n}}$. Controlling switches S1 and S2 switches between single-stage and two-stage amplification. Two-stage amplification provides a smaller temperature measurement range but higher accuracy. In this design, Rs is valued at 1 kΩ. The output voltage for single-stage amplification is as follows:

$$V_{out}={(1+R_1)·V}_{ref1}{-R_1·V}_{ref0}+{R_1·V}_D$$

For two-stage amplification, the output voltage is as follows:

$$V_{out}={(1+R_2)·V}_{ref2}{-R_2·(1+R_1)·V}_{ref1}+{{(R}_1·R_2)·V}_{ref1}-{R_1·R_2·V}_D$$

Adjusting the high-precision resistor S_R1 changes the amplification factor. Adjusting the LT3024 reference voltage modifies the offset voltage, allowing flexible matching of the next circuit stage’s input voltage range based on the thermal test chip’s performance parameters and the experiment’s temperature range. Simple voltage and resistance measurements determine the relationship between temperature and output voltage.

The current signal from the temperature sensor is converted into a voltage signal through a resistor, then amplified by an operational amplifier. The amplified positive voltage of the diode is then fed into the analog-to-digital converter (ADC). The system uses the ADS822, a high-speed 10-bit pipeline ADC, to convert the analog signal to a digital one. The ADS822 performs analog-to-digital conversion through multi-stage processing and parallel operation, delivering a complete 10-bit result each clock cycle while ensuring high accuracy and speed.

First, the input signal is processed, and the output from the operational amplifier enters the sample-and-hold circuit. This circuit samples the input signal every clock cycle and holds the value briefly to ensure stability during conversion.

The signal then undergoes multi-stage processing. The ADS822 has multiple stages, each containing a sub-ADC, sub-DAC, subtractor, and residue amplifier. The first stage processes the higher bits of the signal, generating an approximate digital output and calculating the residue. The residue is amplified and sent to the next stage for further processing.

Next, row and column readouts from the ADC are performed. The STM32F407 microcontroller selects specific channels by controlling the row and column address signals (Row_address, Col_address). The input voltage range is adjusted by controlling the AD_RESL pin voltage. When AD_RESL is high, the input voltage range is 2 V, with a minimum resolution of 1.95 mV. When AD_RESL is low, the input voltage range is 1 V, with a minimum resolution of 0.98 mV.

Finally, for digital output, the STM32F407 microcontroller controls the ADS822 clock via the CLK signal and reads the converted digital signals (AD_bit0-9).

2.2. Microcontroller Design

The microcontroller unit (MCU) communicates with the host computer and controls the acquisition circuit’s working sequence. This involves providing gating address signals for the DG406 multiplexer, reading ADC output signals, and supplying the ADC’s working clock. Figure 3 illustrates the MCU workflow.

To accommodate different thermal testing requirements, the MCU’s scanning address, sampling interval, and sampling duration are configurable. Upon receiving the scanning address signal from the host computer, the MCU generates the corresponding address array internally. After each data collection cycle, the next group of address signals is written into the output register of the corresponding I/O port to achieve channel switching once the acquisition circuit starts working. Adjusting the timer count register sets the sampling interval T and the number of samples N. Controlling the I/O pins for reading and switching address signals ensures that data acquisition and transmission do not affect the MCU’s speed, maintaining the stability and accuracy of temperature measurements.

To meet transient testing needs and achieve millisecond-level monitoring of the thermal test chip, data transmission between the MCU and the host computer uses the USB 2.0 protocol, supporting a maximum speed of 12 Mbps. USB is a widely used protocol standard, offering strong compatibility with most devices. The STM32F407 microcontroller includes a USB OTG FS core and a USB OTG library, making it suitable for USB communication applications. It also includes multiple sample programs, and the library files offer suitable classes for various applications. The HID (human interface device) class, commonly used for peripherals like mice, suits communication applications requiring small data volumes and high-frequency access, meeting the thermal test chip readout system’s needs. Since HID devices do not support receiving commands from the host, additional communication functions must be developed after importing the library files.

2.3. Host Computer Software Design

The system software is developed using the VS Code editor in Python and is primarily responsible for providing interface programs, MCU communication, data storage, and real-time display. The pyusb library facilitates communication between the upper computer and MCU via USB, establishing a connection using the device’s PID and VID.

Upon receiving the output signal, the upper computer program first stores the raw data in a .txt format. Once the output signal is received, the upper computer stores the raw data in .txt format. With a constant forward bias current, a linear relationship exists between the diode’s forward voltage and temperature. Therefore, the upper computer software performs linear calibration to ensure accurate measurements. The thermal test chip is placed in four different temperature environments within a thermal cycling chamber. Once the temperatures stabilize, data from all points are collected. After measurement, linear fitting is applied to obtain the calibration coefficient for each diode. The signal is processed based on the acquisition circuit parameters, converting it into temperature values. Finally, using Python’s PyQt5 graphical interface, the system displays the overall temperature distribution on the chip and the temperature change curve at each point in the window. Figure 4 shows the software interface.

3. Experimental Testing and Results

3.1. Parameters and Verification of the Thermal Test Chip

The thermal test chip used in the experiments consists of a 4 × 4 array of temperature-sensitive diode units, with each unit measuring 1 mm² and offering a high thermal sensitivity of 2.6 mV/K. It is mounted on a 5 × 5 mm² PCB test board using surface-mount technology. The chip’s design integrates heating resistors and sensing diodes, enabling simultaneous heating and temperature measurement capabilities. This setup facilitates precise thermal emulation for packaging and dissipation analysis. Figure 5 illustrates the chip structure, emphasizing the integration of functional elements for high accuracy and fast response times. The PCB test board measures 4.9 × 4.9 cm², is 2 mm thick, and is connected via wire bonding. Figure 5 shows the thermal test chip and its structure, with basic performance parameters listed in

Table 1. Thermal test chip parameters.

Name	Parameter Value
Heating resistance value	11–14 Ω
Diode forward voltage	1.65–1.7 V@10 μA
Temperature sensitivity of the forward conduction voltage	2.6 mV/K@10 μA
Inverse saturation current	10–20 pA
Diode reverse breakdown voltage	≈30 V

.

To verify system readout accuracy, the diode voltage at different temperatures was measured using both the Agilent B1500A semiconductor device analyzer and the designed multi-channel readout system. The temperature range was from 30 °C to 100 °C, with intervals of 10 °C. The data were processed and linearly fitted; the test results are shown in Figure 6. The fitted line for the multi-channel readout system was

$$\mathrm{V}=-0.002542\times \mathrm{T}+1.7392$$

R² = 0.9978, while the B1500A’s fitted curve was

$$\mathrm{V}=-0.002525\times \mathrm{T}+1.7186$$

R² = 0.9997. The temperature measurement percentage error between the multi-channel readout system and the semiconductor parameter analyzer was about 0.67%, indicating good consistency.

3.2. Steady-State Temperature Distribution Testing

Before conducting steady-state temperature distribution testing on the thermal test chip, the readout system was set to use single-stage amplification with a gain of 4.7 and full-chip scanning. The sampling interval was set to 1 ms, with nine samples taken and averaged. Figure 7 illustrates this configuration.

During steady-state temperature distribution testing, the heating unit’s position on the thermal test chip is shown in Figure 8, with a total power of 0.2 W. Temperature data were recorded after heating started, as detailed in

Table 2. Steady-state test data of the thermal test chip.

Temperature Distribution on Chip/(°C)
Heat for one minute	35.9	36.0	36.4	37.3
	36.2	36.1	36.4	38.8
	36.1	36.1	36.2	38.7
	35.5	35.9	36.2	37.6
When thermal equilibrium is reached	40.6	40.7	41.1	41.9
	40.8	40.7	40.7	43.3
	40.7	40.7	40.7	43.4
	40.2	40.4	40.6	42.0

. Based on the collected data, the temperature distribution on the chip was plotted in Figure 9. The experimental data indicate that the readout system can effectively distinguish temperature differences at various points within the thermal test chip, monitor temperature changes, and closely align with the theoretical model. Minor deviations observed in the measurements are likely caused by non-uniform heating across the chip and electrical noise in the data acquisition system. Future designs could mitigate these effects by improving thermal isolation and incorporating advanced filtering techniques in the acquisition circuit.

3.3. Transient Thermal Response Testing

Transient thermal response testing was conducted on the thermal test chip using the same configuration as the steady-state temperature distribution testing. The heating unit’s position is shown in Figure 10, with a thermal power of 0.2 W. Temperature changes at points P1, P2, and P3 were recorded, as shown in Figure 11.

After heating began, temperatures at P1, P2, and P3 rose rapidly. Due to the excellent thermal conductivity of the silicon substrate, the initial temperature difference between P1 and P2, close to the heat source, was minimal. However, P3, at the edge of the chip and farther from the heat source, exhibited a slower initial temperature rise and a noticeable temperature difference compared to P2. As heating continued, the temperature distribution on the chip gradually reached thermal equilibrium. After 15 s of heating, the temperature difference between P1 and P2 increased, while that between P3 and P2 decreased, eventually stabilizing.

The test results demonstrate that in steady-state applications, the thermal test chip readout system can accurately read temperatures, monitor temperature differences at various points, and record temperature changes over time. In transient application testing, the readout system effectively tracks temperature changes and differentiates the thermal response at different points.

4. Conclusions

This study presents a multi-channel temperature measurement system based on a diode array structure, designed to address the precise thermal testing requirements of chips under both steady-state and transient conditions. The system achieves a high measurement accuracy of ±0.1 °C and a millisecond-level sampling rate, and supports diode arrays up to 48 × 48, offering excellent scalability and adaptability for complex thermal environments. By integrating adaptive amplification circuits, high-speed ADCs, and Python-based software with an intuitive graphical interface, it enables real-time monitoring, efficient data processing, and visualization of thermal distribution. Experimental validation demonstrated its reliable performance, with a measurement error of approximately 0.67% compared to a semiconductor parameter analyzer. While minor deviations caused by non-uniform heating and electrical noise were observed, future improvements in thermal isolation and noise filtering are expected to further enhance its accuracy. Overall, the proposed system provides a robust and practical solution for thermal design and evaluation in chip packaging, microsystems, and high-power density applications.