What killed the op amps?

Locating the source of electrical overstress (EOS) events that have damaged semiconductor devices proves challenging, and the challenge becomes even greater when the EOS events don't occur regularly. To illustrate how you can track down EOS events, I'll explain how we helped a customer identify the cause of failure in two operational amplifiers (op amps) (Ref. 1). The first op amp, OP1, experienced a large EOS event that destroyed many parts of its circuit. In the second op amp, OP2, an EOS event damaged only a thin-film resistor.

Both op amps had one or more pins that failed continuity tests—the first sign of a failed device. In addition, OP1 exhibited degraded operation at several other pins. Neither OP1 nor OP2 passed functional testing. After we ran electrical tests, we opened each op amp package to inspect the failed device. Visually comparing a failing device with a working device helps you locate failure sites.

In contrast, the OP2 device showed none of the classical signs of EOS—disturbed metal traces and burn marks. Only a suspect resistor (Figure 3) showed a slightly different color. At this point, we might have concluded an EOS event did not cause the failure. Perhaps the NiCr resistor underwent oxidation or corrosion instead. But resistors in the same area showed no similar signs of discoloration, and it's unlikely a wafer-fab problem would have affected a single resistor. Also, we observed no corrosion by-products, nor did we find a defect in the passivating oxide that would have given corrosive chemicals access to the resistor.

We observed that the open resistor (Figure 3)—connected to the negative input in OP2—caused the device to operate improperly. When this resistor "opened," it cut a feedback path and caused the op amp's output to swing high and stay high regardless of the applied input signal. Probing the damaged resistor showed a normal trace on the input side, an indication the EOS event did not damage other circuits in the op amp's input path.

Track down causes

Once we knew about the damage that caused each failure, we had to track down the causes. The first step in identifying the cause of an EOS event involves getting information from the person who reported the failures. We needed to know which circuit and board configurations were in use when the failure occurred, what the test conditions were when the part was last known to work properly, and what events took place after the last test or use during which the part functioned properly. A schematic of the circuit that incorporated each op amp showed the connections between the op amp and all other components and "outside world" signals.

We looked for patterns in the damage observed on each op amp. These patterns, and an understanding of the circuit components that surrounded each op amp, yielded information about the source and intensity of EOS events. External signals that pass through large impedances, for example, are less likely to provide a source of EOS energy. An impedance diminishes current flow and provides some protection. Direct connections to an op amp from power supplies and other devices' pins may offer low impedances and, thus, better conduct EOS energy to a semiconductor device.

Figure 1 Damage in op amp OP1 occurred in many areas, leading to the conclusion that a high-energy event caused the op amp to fail.             Figure 2 The fused output line in an op amp shows the damage caused by an EOS event. In this case, the fusing opened the connection.             Figure 3 Mild damage to a different type of op amp (OP2) destroyed only a single resistor.

The circuit that included OP1 used the device as a unity-gain noninverting amplifier with the output connected to a cable harness located off the circuit board. In this configuration, the op amp's output connected directly to its negative input. The input signal to the amplifier connected directly to OP1's positive input from a source on the board.

Based on our observations and on the application of the op amp, we believed damage occurred because of a positive voltage (with respect to V+) applied to the op amp's output pin. A partial schematic for OP1 (Figure 4) shows the path of the current flow from the op amp's output pin through Q70 and Q75 to the V+ line. Q70, a large output transistor, could handle the power in the EOS event, but Q75 could not, as indicated by the aluminum "short" we found at the base-emitter junction of Q75. This small transistor could not dissipate the large amount of energy in the EOS event without shorting. After the current reached a critical level, the metal output line that went to the bond pad fused, and a large section of the line blew away as shown previously in Figure 2. To blow such a large section of metal requires a massive current pulse (1 to 2 A) of short duration.

Other damage occurred in OP1, too. When the metal output line opened, the current quickly decreased to 0 and voltage increased rapidly. Because the op amp's output connected directly to its negative input, the EOS voltage pulse caused damage that was observed around the –Input connection on the op amp (Figure 1). In my opinion, a parasitic inductance in the source of the EOS signal caused the output voltage to increase rapidly.

OP2 showed less damage than OP1—only an open NiCr resistor—which made it difficult to determine what caused the device to fail. Electrical tests demonstrated that other elements connected to the NiCr resistor functioned correctly. The resistor is connected between the pad and the input stage. The lowest breakdown path existed for a positive voltage from the pad to the negative supply. Other circuit damage would have occurred if the charge had taken a different path. Thus, the EOS pulse of energy must have entered the negative input pin.

The lowest resistance breakdown path existed between the negative input and the negative-supply rail, so the EOS current traveled this route. But because we saw no damage to the metal line or elements other than the resistor, we concluded this EOS event produced only a small amount of energy. Also, a slow pulse would have damaged the center of the NiCr resistor instead of damaging the entire resistor area. So, we assumed the EOS event occurred very fast and had a fast rise time.

The next stage in our hunt involved running experiments to try to duplicate the failures. We had to make some assumptions about the type of EOS event that caused the damage. For example, we assumed a test lead would supply sufficient inductance (~2µH) to cause a voltage spike, so we didn't place additional inductance in our test circuit. And we had to make some guesses about voltage and current levels, energy delivered to the circuit, and the duration EOS of events.

To test OP1 devices, we used a Tektronix curve tracer to supply a 25-V pulse with a duration that ranged from 10 to 50 ms. A 3-ft test lead connected the curve tracer to the DUT. Under these conditions, the test part did not fail in the manner we had observed in the failed OP1 device. A second attempt with the voltage set to 350 V and with a series resistor that limited peak current to 2.5 A produced damage similar to that seen in OP1. The pulse damaged the same areas of the circuit damaged in OP1, but we observed more severe damage in the test part. A reduction in the voltage level or a higher series resistance probably would have decreased the amount of damage, but we felt we had found the cause of damage, so we didn't perform additional experiments.

Figure 4 A partial schematic of the OP1 op amp shows the paths taken by current during an EOS event.

Using our test results, the customer located the probable cause of the failure—an ungrounded test cable at the test station where the failures were discovered. The ungrounded cable could charge to significant voltages, and when connected to the board, it would discharge into the board's circuits and damage the op amp and other components.

The source of failure in OP2 proved more difficult to locate. First, we applied a voltage to negative input on a test device and increased this voltage until the op amp's input resistor opened. A +17-V signal on the op amp's negative input caused the resistor to fuse, but it appeared different from the failed resistor in OP2.

Rather than showing a failure throughout the resistor, the resistor in the test device showed a line across the resistor. We decided to apply more energy to completely fuse the resistor and to apply the energy quickly to prevent heat loss from the resistor.

A curve tracer provided too slow a pulse to rapidly heat the entire resistor, so we tried a transmission-line pulse (TLP) tester. This type of tester charges a length of coaxial cable to a preset voltage and then discharges the cable into a DUT. A TLP tester can produce a rectangular current pulse with a rise time <2 ns and a variable pulse width. When we charged the cable to 250 V, it produced a peak current of 0.5 A, which fused the resistor in the op amp in 55 ns. The result of that pulse test matched the damage seen in OP2.

This result doesn't imply that charge from a cable assembly damaged the part. But it does imply a fast pulse with a quick rise time, and a current of about 0.5 A would produce similar damage. Further work by the customer tracked a possible cause to an oscilloscope close to boards undergoing testing.

The customer found the scope radiated a high-energy electric field and thus induced a charge on nearby parts. When technicians touched test instruments to the boards, a discharge occurred. Proper shielding and removal of charge eliminated the problem of failed op amps on boards undergoing testing.

In some cases, a failure analyst just doesn't know enough about a failed part to understand how it operates. The vendor knows the most about the part, but the customer knows the most about the part's actual application. Thus, both parties must freely share information to solve tough EOS-related problems.

It's easy for a vendor to blame EOS conditions on a customer. In response, a customer may claim to have all the needed EOS safeguards to prevent overstressing parts. This sort of "blaming the other party" never gets to the root cause of failures caused by EOS events. The customer and vendor must work together to determine what occurred and to correct the problems.