Not a Success Story

Home page....: Frigus Primore home page
Calculator.....: Thermal territory for a heat source

---

Figure 1
Thermal territory for the communication device.


Introduction

It is always nice to read a success story but from a professional point of view I find failure stories more interesting. They are often truly sad but they do push things forward. Experience is also to a large extent based on mistakes made. A thermal designer that has a large stack of failure stories is therefore well equipped for the job.

Speaking in very general terms there are thousands of ways to do things wrong but only a few ways to them right. From that perspective it is surprising that not more mistakes are made. It is nevertheless inevitable that things go wrong now and then and at those occasions it is important to learn as much as possible about the reasons.

As any other professional group thermal designers also get their share of failures. One reason why they not are told more frequently is that they often are embarrassing both from the personal point of view and from that of the company. Another reason is that one tends not to make the same mistake twice and certainly not three times. Repeated mistakes of the same kind are therefore exceptional and the ones that remain are consequently difficult to relate to as typical.

What might appear a bit strange in this context is that I never have seen a case where poor calculation accuracy has been the cause of an important failure. I do not deny their existence because I have seen many calculations that have been terribly wrong but the error was in all those cases detected so soon that it did not cause much harm. The idea that calculation accuracy is the key to successful thermal design is therefore not at all in line with my experience. Since I do not want to be misinterpreted on this pointI do stress that it is important to make accurate calculations but only to the degree that they reasonably well match all other uncertainties. Quite a few of the failures that I have seen have however been caused by a total neglect of thermal considerations, particularly in the early phases of the design process.

A very apparent example of a thermal failure is when a system dies while exposed to extreme temperatures. I have seen half a dozen cases. They are always dramatic because they appear unexpected and they also tend to create panic in the sales department. These failures are either caused by extreme component temperatures or by poor component quality. I have heard several stories about the former but I have only experienced the latter. Somewhat surprisingly a few of those were actually provoked by temperatures at the low end of the scale.

To prevent system failures is the classical argument for thermal design. It has however lost much of its bite as the numbers of incidents, for many different reasons, have declined. It is therefore high time to focus on what thermal design can do to make the design process smooth and in particular, how to avoid costly time delays and redesigns. The problems that appear in the design process form a very heterogeneous group. There are nevertheless a couple of common denominators: they are always discovered too late and they are often caused by an unfortunate combination of several mistakes. The story that follows is a typical example.


The problem

A system often consists of several PCBs that need to communicate. As far as an amateur in electronics understands it, this communication is always problematic but there are several electrical tricks that can facilitate the matter. In a particular project it was subsequently decided to create a custom design circuit that substantially would ease the inter talks. The solution apparently had many advantages but one disadvantage was that the device had to be placed on every PCB in the system. To speed up the design process it was further decided to run the design of the custom circuit and the PCBs simultaneously. The thermal aspects were, for reasons unknown, forgotten at the project launch.

The project leader consulted me some weeks later. He had a diffuse feeling that the thermal issues had to be considered. At that time the project had advanced to the point that a decision had been taken on the package for the custom circuit. A heat dissipation estimate was also at hand, 3 W.

Given just the data for the component it is of coarse impossible to make any reasonable thermal prediction. With a bit of additional information about the PCBs and the air velocity one can however produce an interpretable estimate. A method that is convenient for the discussed case is the thermal territory method. A thermal territory can shortly be described as the smallest rectangular surface of a PCB that a component needs for its cooling. Those who not are familiar with the method can try a simplified calculation procedure in the link provided at the beginning of this article.

My first insight into the matter resulted in an image that looked something like figure 1. A more precise assessment would have been a great help at this point but it could not be produced on the bases of the data available. A swift glance at the image was in this particular case sufficient to realise that it would be difficult to keep other heat dissipating components away from the thermal territory of the custom circuit. Both the project leader and myself therefore agreed that we were dealing with a grey zone case.

What we did not agree on, was how to deal with the problem. The project leader argued that the difficulty could be managed if everything was done to focus on it in the layout phase. I argued that the only safe solution was to prepare for the eventuality that a heat sink should become necessary. The reason for this difference of opinion soon became apparent to me.




Figure 2
Cavity up and cavity down packages favour different heat flow paths.


Cavity up or cavity down

The package that had been selected was of the ceramic pad array type with a cavity facing upwards. This kind of package strongly favours the heat flow path from the chip to the PCB and does not at all perform well with a heat sink, figure 2. An alternative could have been a cavity down type of package for which these tendencies are the reverse.

To change the package at this phase was however not that easy. The signal configuration of the pads had already been decided and changing the package would essentially mirror this configuration. Since several of the PCB projects already were working on their preliminary layouts, such a change would inevitably result in some redesign and consequent time losses.

After having consulted the PCB designers it was decided not to make any changes. I had done my best to inform about the problem and was in addition sure that I had been understood. There was not much else to do than to accept the decision.


An additional problem

The project leader came to see me several months later. An additional problem had appeared. Measurements had detected that the maximum junction operation temperature for the custom circuit was 85 degC and not 100 degC as had been specified. It is apparent that a design which already is on the margin to what can be done not can put up with such a radical decline of a thermal property.

I nevertheless made several efforts to save the project with a heat sink. In vain, the junction to case thermal resistance was just to large. It could not be done. What I could detect however, was that a cavity down package would have provided the margin need to deal with the set back.


The consequences

The incident caused a 3 week project delay. It is true that the major cause was a mistake made on the chip level design and that it was difficult to predict. The second mistake however, was not to use thermal design from the project start. One can only speculate about the outcome if this had been done but one possibility is that it would have been different.

It is not easy to calculate the cost for an incident like this. If one disregards the market costs and assumes that 20 persons were involved it is nevertheless possible to make a coarse estimate. It indicates a cost that is 50 times the cost to get started with thermal territory calculations, software and training included. Another way to express this ratio is that if an incident of this kind can be avoided every 5 years, the pay back time for the software is of the order 5 weeks. This is 30 times better than what usually is regarded as a good investment. It is thus possible to lower the assumptions in this cost analysis with a factor 10 and yet get a very favourable outcome for a front-end thermal design software investment.

This story did unfortunately not have a happy end. It does however show that it is important to practise front-end thermal design methods, however simple or primitive they may seem.


Ake Malhammar