This chapter is intended to give some pointers towards finding faults in a microprocessor-based system. This chapter is firmly based on experience and could equally well have been entitled, ‘Mistakes I have made’.
The whole process of fault-finding should be undertaken slowly and carefully. There is a popular misconception that you have to keep busy, taking measurements, making adjustments and changing components. But, in fact, most of the time is spent just sitting and thinking (don’t forget the last two words!).
Collect the symptoms and write them down. Be wary of other people’s idea of the symptoms. If they have misunderstood what is happening you could waste hours or days going off at a tangent. If you forget to write them down, then sooner or later you will be back repeating the same checks.
In most cases, a piece of equipment or a circuit fails due to a single fault. Two simultaneous but unconnected faults are very rare. There are two popular ways of converting a small problem into a large one. These are static electricity and plugs etc.
Static electricity
When two different materials rub against each other, some negative electrons tend to migrate from one material to the other. This results in a voltage difference between the two materials. The amount of voltages can be very high – several thousand volts. If we walk across a carpet or sit on a plastic covered chair, we can become lethal to an integrated circuit designed for 5 V. Many integrated circuits have antistatic precautions built in but they have limited success. There is a trade-off here in that the better we make the antistatic precautions, the slower the integrated circuit can switch.
We can overcome the problem by reducing the build up of static by allowing it to leak away. In carpets, clothes and furniture we can do this by adding a wax or polish that absorbs and holds a small quantity of moisture. A slight dampness is a very effective way of preventing static problems. For this reason, the weather and air humidity is important. The death rate of integrated circuits tends to vary seasonally! It is not helped by air-conditioned plant where the humidity is low. The effect of static electricity on integrated circuits is difficult to predict. It generally causes small localized failures which can have very peculiar effects.
Better than spraying ourselves with water, we can take a more hightech approach but how far to go in this direction depends on what is at stake. If we are going to handle a couple of cheap AND gates once a week, then only the simplest precaution is worthwhile. However, sitting on a production line plugging in microprocessors will make any precautions economic.
The simplest method is to have a conducting band clipped around your wrist with a lead going off to a ground (earthed) point. These wristbands are made of rubber into which carbon has been amalgamated to allow it to conduct slightly. As well as the wristband we can place a sheet of this rubber on the bench top and ground the bench. Such antistatic workstations are very effective. A word of warning. Do not make your own wrist strap from a length of copper wire. This offers a very low resistance and provides no protection against electrocution in the event of accidentally touching a power line.
At home, just avoid working on a plastic table or chair or wearing clothing made from man-made fibres. Natural materials like cotton, wool and untreated wood naturally absorb some water and are fairly safe. A nice wooden bench coated with polyurethane varnish is effectively a plastic bench and should be avoided.
Problems with plugs
Many plugs used between pieces of equipment have a large number of pins. Pulling one of these out with the power connected is going to disconnect some voltages before others. This can prove fatal for integrated circuits. Either all the supplies must be on, or all should be off so never plug or unplug anything with the power on. For the same reason, never remove or replace an integrated circuit with the power on.
Are the power supplies turned on? Do you need two supplies? If you are using two supplies, are they connected together to keep their voltages in step with one another? If a ground connection is required, is it connected?
Most power supplies have floating outputs. That means that a 5 V supply, for example, will have a 5 V difference between its two terminals but neither is connected to the ground potential. This means that if we connect the negative terminal to earth, as in Figure 18.1(a), the other terminal goes to +5 V. If, on the other hand, we make the connection shown in Figure 18.1(b), the other terminal will become –5 V.
Figure 18.1 Connecting floating supplies
Have a look at the soldering if it is visible. It should be smooth and shiny. Any dull and craggy looking areas are suspect. If the integrated circuits are plugged into bases rather than being soldered, have a look to see if they have been inserted the right way round. Unfortunately, integrated circuit manufacturers take few precautions to prevent this type of error.
In most integrated circuits, the pins are numbered around the outside as shown in Figure 18.2. The position of pin 1 is always on the left-hand side of the end which has an indentation when viewed from the top as in Figure 18.2. When looking for the indentation don’t be mislead by a small circular mark where the plastic has been molded. The printed circuit board usually has either a number ‘1’ or a small square or other mark to indicate the position of the first pin.
Figure 18.2 Pin numbering of ‘dual in line’ (DIL) chips
Figure 18.3 shows the pin grid array (PGA) layout. Notice that the letters skip from H to J because of the possible confusion between I and 1. The device determines the number of pins. The one shown happens to be the elderly Intel 80386. The Pentium has 21 pins along each side.
Figure 18.3 Pin numbering of Pin Grid Array (PGA)
Apart from the standard voltmeter and an oscilloscope the only other simple piece of gear that may be helpful is the logic probe. It is better than the average oscilloscope at detecting very short voltage spikes and is faster to work with than a voltmeter.
Logic probe
A logic probe is a simple instrument that has two power connections and the other is a conducting tip that can be touched on points of interest. The general layout is shown in Figure 18.4. There are three LEDs on it. The first two show the logic states 0 or 1 and the third one indicates the presence of a high frequency square-wave or a single, very short duration, pulse, called a ‘glitch’.
Figure 18.4 A logic probe
Simple tests to make with these pieces of test gear
We can check some of the voltages on the microprocessor pins. If possible, it is a good idea to check on the actual pins rather than the base into which it is plugged. Doing this ensures that the base connections are also OK. It would also find the bent pin shown in Figure 18.5.
Figure 18.5 Pin bending not recommended
The likely pins that are worth checking are the ones carrying a dc voltage like the power supplies and the interrupts. It is worth keeping an eye on pins that should be at 0 V. When using a voltmeter, they can sometimes show 0 V when they are disconnected and floating. If you use your voltmeter to measure the voltage between the positive supply voltage and the suspect pin, it will still indicate 0 V showing that something is clearly amiss. A logic probe would not be fooled by a floating ‘zero’, it will not show a logic zero if it is floating.
The next job is to see if the microprocessor is running at all. We can do this by using the oscilloscope on a clock signal. Assuming that the clock signal is OK, we must next check that the microprocessor can follow an instruction and that the address and data bus are being read correctly.
A good check on the operation of the microprocessor can be arranged by getting it to do a simple repetitive program consisting of a permanent ‘no operation’ code. A no-operation code will instruct the microprocessor to do nothing except read the next instruction from the data bus by simply incrementing the value on the address bus. This new instruction will be another no-operation code and so the address bus will be continuously incremented. To provide a permanent no-op input we can solder or otherwise connect the required logic codes to the data bus. This is called hard-wiring the data bus. As the address bus counts up in binary the lowest address line will be switching rapidly between zero and one giving a square wave output.
If we look at Figure 18.6, we can see that line A1 is running at half the frequency of line A0. Similarly address line A2 has half the frequency of A1 and so on all the way along the address pins. If we connect an oscilloscope to each line in turn, the frequency should reduce steadily. Check for the halving of frequency on each address line and errors in wiring like short circuits between address lines will become apparent.
Figure 18.6 The address bus counts up in binary
If we get this far and still things seem wrong, we are into serious faultfinding.
All the previous pieces of test gear have failed when we try to see what is happening on the address and data buses under real operating conditions. The oscilloscope cannot watch more than two different places at the same time but we may need to monitor a larger number, perhaps 50 or more places and then slowly check the information back or print it out. An instrument called a logic analyser can achieve all these functions and much more.
It can answer such questions as:
• What values actually appear on the address bus when we cause an interrupt to occur?
• Is the correct program actually being run?
• Are there any unwanted voltage spikes occurring?
The design of a logic analyser is basically a very simple combination of shift registers. You may remember we looked at shift registers in Chapter 6. The register was loaded with data and, on each clock pulse, the data is moved one place to the left or right as required. Now imagine a shift-right register that can hold 36 bits of data. If we connect it to A0, the first line of the address bus, and run a program, the logic values of that address line will be copied onto the shift register, pass along to the end with each clock pulse and eventually start to fall out the far end (see Figure 18.7).
Figure 18.7
Now if we had four such registers, we could collect data from any four parts of a circuit at the same time. For example, we could monitor the lowest four address lines, which would be called A0, A1, A2 and A3.
In the centre of the register is a window. This means that we can access the centre of the shift register at this point to read off the data and to make comparisons. In Figure 18.8 only the four registers are shown for clarity. A simple arrangement like this would be referred to as a 4×16 (four by sixteen) logic analyser. In logic analyser specifications, the number of registers would be spoken of as the number of channels. In real life, we would never find such a simple arrangement of registers. Logic analysers could contain, say, 80 channels, each containing 4096-bit shift registers. This would be referred to as a 80×4 kbyte logic analyser. With one this size, we could monitor any 80 different points on a microprocessor-based system. Which points we choose are up to us, we could choose the whole of the address bus and the data bus and some control signals or any other points of interest. The choice is entirely ours.
If, in Figure 18.8, the four registers were being used to monitor four address lines, we may be suspicious of the line showing a constant value of logic 1. This may indicate that this line has become short-circuited to a positive power supply, or be disconnected and is floating high. Don’t leap off your chair in excitement though – this is only one explanation. It could happen for these reasons or it could be running a part of the program where this would be the expected result.
Figure 18.8 Four shift registers can make a simple logic analyser
So what about the window?
In the window, the logic analyser will ‘see’ a bit of data from each of the channels. We can load the combination that we are searching for. For convenience, we enter the values in hex numbers and as the clock pulses arrive from the microprocessor, the data moves across and is continuously compared with the number we have entered. When a match is found, the clock is switched off and the data is ‘captured’. We can now move backwards and forwards along the registers and see the operation of the microprocessor ‘frozen’ in time. The benefit of positioning the window in the centre of the shift register is that it allows us to observe the program action before, as well as after, the chosen moment.
Extra facilities
A ‘real’ logic analyser has some extra facilities, like performing the capture not on the first time our input is seen but after perhaps the 200th occasion to take care of repetitive loops in the program. They also allow a ‘don’t care’ condition on the inputs so in the window of a 20-channel logic analyser we could enter the hex code 7XXX2. This would perform a capture on any data that starts with 7H (01112) and ends with 2H (00102).
A glitch is a very short duration pulse that can occur in logic circuits, either from external interference or as a result of poor design. They can cause unwanted switching in the logic circuit and cause the microprocessor program to crash. They are exceedingly short, just a few nanoseconds and this makes spotting them very difficult. They are usually too fast for an oscilloscope but some logic probes have a ‘glitch-catcher’ built in, but they can only tell us that a glitch has occurred, not when it occurred. This is the information that will be needed if we are to track down a design problem.
A logic analyser may miss it because the incoming data is sampled once per pulse and if it misses the glitch it will not be recorded. To overcome this, the logic analyser can use its own internal clock that is running much faster than the system clock so a single logic one may extend for 10 or 20 bits in the register and a glitch may well be recorded. Figure 18.9 shows the internal clock running at 10 times the microprocessor clock. Some logic analysers have a built-in glitch catcher and use it to capture the correct section of data. As we can see, the logic analyser is a very useful and sophisticated piece of kit. Using it, however, is a slow process. There are lots of connections to be made to the circuit and much sitting and thinking.
Figure 18.9 Glitch catching
In each case, choose the best option.
1 Damage due to static electricity:
(a) can only occur in the winter.
(b) is best prevented by wearing wet clothing.
(c) is only possible in carpeted areas.
(d) can be reduced by wearing a grounded wrist strap.
2 If the two power supplies were connected as in Figure 18.10, the result would be:
(a) smoke pouring from both power supplies.
(b) an output of +5 V but twice as much current.
(c) an output of +10 V.
(d) no output but no smoke either.
Figure 18.10
3 The arrow in Figure 18.11 indicates pin:
(a) 2.
(b) 8.
(c) 11.
(d) 16.
Figure 18.11
4 A logic probe:
(a) indicates whether your fault-finding technique is based on sound reasoning.
(b) can detect the difference between a disconnection and a grounded connection.
(c) can store a stream of data.
(d) detects the presence of static electricity.
5 A logic analyser quoted as 20 channels × 1024 bits:
(a) will show a four digit hex number in its window.
(b) can monitor any 1024 points at the same time.
(c) would store a total of 1044 bits of data.
(d) can monitor any 20 points at the same time.