4. SEMICONDUCTOR PROCESS CONTROL

4.1. DIE ATTACHMENT EVALUATION

Objective: To evaluate methods for detecting poor die attachment in semiconductor devices with initial emphasis on the determination of the applicability of thermal measurements to this problem.

Progress: Modifications of the revised TRUMP [1] thermal analysis computer program for use on the NBS computer were completed. Modeling of the various structures to be analyzed has been undertaken. Initial computer runs using both a simplified mesa diode on a TO-5 header and a round silicon chip on an infinite heat sink indicate junction temperatures that are in excess of those that would reasonably be expected. A more detailed study of the modeling and modeling input procedure has been undertaken. (R. L. Gladhill)

The series of measurements of steady-state and transient thermal response on transistors was completed. Measurements were made on transistors with controlled voids that were approximately 15 and 25 percent of the 35-mil (0.89-mm) square chip bonding area. The voids were formed by ultrasonically machining dimples, 15 and 20 mils in diameter, respectively, into the bonding surface of the TO-5 header. Measurements on transistors with 40-percent void areas were reported previously (NBS Tech. Note 773. pp, 22-24).

For the steady-state measurements, the heating current and voltage were 50 mA and 10 V, respectively. The transient measurements were made 10 μs after the termination of a 5-ms wide power pulse; the heating current and voltage were 150 mA and 10 V, respectively. The spread in thermal response of the devices with the 15-percent void areas was found to be significantly larger than in that of the other devices. Although two groups of transistors were bonded with 15-percent void areas, both thermal response measurements and radiographs indicated that there was poor bonding in the region around the dimples in all but five of these devices from one of the groups. For comparison, nine control devices without voids bonded at the same time were also measured. Nine control devices without voids were also measured with the group of nine devices with 25-percent void areas.

The average percent increase in steady-state and transient thermal response of the devices with voids over their respective controls is shown as a function of percent void area in figure 11. The data presented indicate that, at least for this particular device, measurement of steady-state thermal response (or thermal resistance) would not be an acceptable industrial screening technique to cull out devices with various size voids due to the low sensitivity of the technique, while the transient

thermal response for a particular heating power pulse can discriminate between devices with and without voids.

Measurements of transient and steady-state thermal response were also made on ten commercial transistors that used the same 35-mil (0.89-mm) square transistor chip that was used in the in-house bonded devices. The commercially bonded chips were in solid metal TO-5 cans while the in-house bonded chips were attached to standard iron-nickelcobalt alloy, glass-backed TO-5 headers. Although the thermal resistance of the two classes of devices is quite different, the thermal resistance of the in-house, glassbacked, devices being approximately 60°C/W and that of the solid metal encased devices being approximately 40°C/W, the average junction-to-case temperature difference for the 5 ms heating power pulse in the commercial devices was 20.3°C (excluding one poorly bonded device), and that for the controls for the in-house bonded devices was 21.3°C. The sample standard deviation in both cases was 0.4°C. Thus, the in-house control of the die-bonding process appears to be of the same order as for the commercially bonded devices.

Long-term, single-operator measurements of steady-state and transient thermal response were initiated to check the repeatability that can be achieved with the transistor die attachment evaluation equipment. Thus far, 12 sets of measurements have been made on two controls and two transistors with 40-percent void areas. Repeatability measurements of transient thermal response, for a heating power pulse of 5 ms, are also being made with the device mounted in a standard socket instead of in the temperature controlled heat sink. (F. F. Oettinger and R. L. Gladhill)

DIE ATTACHMENT EVALUATION

Specimens are being prepared for an experimental study of the relationship between thermal response and voids in power transistor die attachment. The study calls for three groups of 24 devices. Each group consists of 12 n-p-n silicon power transistors, 60 mils (1.52 mm) square, bonded to steel, with gold plated molybdenum pad, TO-66 headers with 26-, 34-, or 43-mil (0.66-, 0.86-, or 1.09-mm) diameter dimples ultrasonically machined into the bonding surface to produce voids that are approximately 15, 25, or 40 percent of the total chip bonding area, respectively, and 12 control devices without voids. (T. F. Leedy and J. Krawczyk)

Plans: The long-term, single-operator measurements of steady-state and transient thermal response will be completed. Upon completion of the bonding of power transistors on TO-66 headers with various size controlled voids, measurements of thermal response will be made on the transistors for evaluation of the die adhesion. The analysis of heat flow to determine the limitations of thermal response techniques for detecting poor die adhesion in the diodes previously investigated will resume.

4.2. WIRE BOND EVALUATION

Objective: To survey and evaluate methods for characterizing wire bond systems in semiconductor devices, and where necessary, to improve existing methods or develop new methods in order to detect more reliably those bonds which will eventually fail.

Progress: The experimental phase of the investigation of the effects of geometrical variables on the destructive, double bond pull test was completed. Pull strength was measured as a function of loop height, pull angle a, pull angle ß, and position of pulling hook for two-level bond pairs. These measurements yielded results which agreed with theoretical predictions. The vibration modes of two unconventional ultrasonic bonding tools were studied during bonding. One was significantly different from conventional tools.

Pull Test Evaluation Measurements of pull strength as a function of bond loop height for two-level bonds were repeated using bonds made on two different bonding machines. Machine A had been used in previous two-level studies, while machine B had not. For bond pairs made on both bonding machines with the first bond either to the high pad or to the low pad, the experimental results were in agreement with theoretical calculations based on resolution of forces (NBS Tech. Note 555, pp. 31-25). Figure 12 shows the data obtained for bonds made on machine B. Results previously reported (NBS Tech. Note 754, pp. 21-23) indicated that agreement between experiment (using machine A) and theory was obtained only for bond pairs where the first bond was made to the high pad. Reexamination of the data has shown that the theoretical analysis applicable to single-level bonds had been incorrectly applied to the data.

WIRE BOND EVALUATION