Cohu Temperature Control 

Background:

Cohu is an industry leader in backend semiconductor chip testing. Cohu uses a device to test these semiconductors known as the Delta MATRiX semiconductor tester. Located inside the tester, a semiconductor handler actively tests the performance of semiconductors at specific temperatures on a 4x8 array, known as a test bed. It does this by passing cooled or heated air through a channel underneath the test bed, maintaining the temperature of the semiconductors placed on the test bed at temperatures from -55° C to 175° C. This wide temperature range is typically seen for semiconductors used in automotive applications. 

This testing is done in a high-speed automated method to maximize the number of devices tested over time and thus reducing the cost of each test. After the test head is placed over the semiconductor in its socket, they are tested at constant temperatures ranging from -55 °C to 175 °C, depending on the application. This temperature is maintained by cycling air through a valve and interconnected channel which passes underneath all the sockets on the testbed. 

Project Objectives:

Due to the large variation in testing temperatures, naturally occurring temperature gradients exist between the sockets during testing. Without a control method in place, these temperature variations between sockets topped +/- 5.6 °C. Even after the current control method discussed in subsequent sections was implemented, temperature variations sometimes exceeded +/- 3.3 °C. The team's goal for this project is to design a control mechanism that can control device temperature through the socket on the test bed shown below.

Specific requirements are listed below.

Design Requirements:

• Design a miniature mechanism (valve) to control the airflow 

• Minimize temperature variation between sockets

• Manage temperature deviation accurately to +/- 2° C across the desired temperature range 

• Automate the temperature control system

What was done to satisfy those requirements:

• Redesigned exhaust manifold:

  • Redirect two vents through a manifold

  • Mounting mechanism to secure PTFE tube to manifold

• Design in-house valve

  • Miniature PTFE/Aluminum butterfly valve

  • Electric motors controlling valve handwheels

• Real-time automated flow control electronics:

  • Arduino platform, including software

  • Motor controllers, including the wirings to the actuators

• Assembly of both mechanical and electronic components

Final Design:

Our final design consists of a couple of components that work together to control flow on an individual socket basis shown above in figures 4, 5 & 6. Located below is a basic listing of all the components that make up the device and we will go more in-depth on these on our Final Design page.

Components

• A Manifold, redesigned to transport air from the two exhaust channels of the socket to a single port 

• 4mm OD PTFE Tubing, fits with Cohu’s in-house compression fittings, is highly flexible, and can withstand the extreme temperatures associated with this application 

• A custom-made Butterfly Valve, manufactured out of PTFE stainless steel and strategically located to limit the distance the air travels from its exit out of the manifold to its entrance to the valve 

• TowerPro MG29B servo motors, selected as the actuating motor for the valve due to its low cost and metal internal components, which would avoid deforming under any potential high heat surrounding the test machine 

• A standard 5mm to 8mm Aluminum Couplings, used to connect the shaft of the motor to the valve

• A custom-made Mounting Bracket, used to hold the entire device in place

• An Arduino Mega 2560, programmed precisely for our application and possessed sufficient hardware (with external components) to measure temperature and control valves for four different sockets

• PT1000 RTD Temperature Sensors, located inside the machine and used to collect temperature readings from the socket

• A MAX31865 Amplifier Board, used to amplify signals received from the temperature sensors to the Arduino

• A 16x2 Character LCD, used to display the status of the system

Summary of Performance Results:

In order to make sure that the device works properly and that it can manage the temperature deviation accurately to +/- 2 °C across the desired temperature range, we decided to test three different control algorithms in order to decide which one worked the best. Although we found that ON/OFF control showed the best results, the incremental method showed similar results with much less motor movement and as a result, will have a longer lifespan. The average root mean square error (RMSE) between the sockets was only a 4.86% increase for incremental and the change in high and low temperatures was also minimal. Incremental control was decided to be implemented into the final design.

After confirming our algorithm, we decided to test the heating of the device through multiple temperatures. As shown in Table 2 and the graphs above, the device achieved the desired performance requirements. During heating tests, the greatest difference from the set temperature was 0.61 °C when testing at 125 °C. The greatest average RMSE was only about 0.16 °C. The error can be seen increasing as the set temperature moved farther away from the ambient temperature as expected at greater temperatures due to the intensity of heating causing more variation. Choosing a smaller increment of movement also led to a decrease in error as movement resolution is increased; however, this led to a longer time to achieve the desired temperature.

For the final tests, we decided to test the cooling of the device through multiple temperatures. Testing data is shown in Table 3 and the graphs above. During cooling, the error was much greater with the largest difference and error of 2.57 °C and 0.5902 °C respectively. The set temperature did not seem to influence error or the distance as all measurements showed similar results. The difference from the set temperature was always greater below the set temperature than above. Decreasing the adjustment rate did not lead to improvements but instead increased the error. There were also significant oscillations that could be attributed to the Delta MATRiX itself. Its method of conducting cooling through the constant mixing of gases has oscillations that the controller does not account for and can not remove. 

Final Presentation: