Martin Bartholomew graduated from the University of Bath in 1973 with a BSc (Hons) in Materials Science. His interest in electronics led him to pursue a career in microelectronic packaging. He has worked for several major semiconductor manufacturing companies including GEC, Plessey, STC and Siliconix, where he was involved in a wide range of development/process/quality engineering aspects of the manufacture of ICs, opto-electronic and power (Mosfet) devices. Since joining TWI, Martin has worked on various projects related to microelectronics, such as the EUREKA EU462 programme for multichip module development, packaging of speech synthesis modules, chip on board technology and use of novel materials for device encapsulation.
Martin Bartholomew discusses the different approaches that can be used to achieve high quality products in a microelectronics manufacturing environment. The main focus of this brief review is use of statistical methods for quality improvement.
The cost of a defective microelectronic device, such as a transistor or IC, can be extraordinarily high. Component failures in space launchers or payloads can cost millions of pounds. The costs of malfunctions in avionics systems are measured in terms of loss of life.
The traditional approach to ensuring production of an absolute minimum of poor quality, unreliable components, has been to put the devices through an extensive series of in-process inspections and tests to remove defectives. Various environmental screening regimes and burn-in procedures have been used to minimise the incidence of 'infant mortalities' in the field. It is much better to pick out the latent failures before the devices are assembled into boards. The cost of rectifying a defective PCB is much higher than that of rejecting a component on the production line.
This typically western approach to quality assurance has served its purpose in manufacture of relatively small quantities of expensive components, aimed mainly at military and aerospace markets.
In Japan, and increasingly in other Far Eastern countries, the approach is quite different. With intense competition in consumer electronic markets, a different quality strategy is necessary, the cost of the mass inspection and test regime being too high. The consumer expects a sophisticated product that performs reliably over its designed lifetime. What is required is an active system for improving quality, rather than the more usual passive one that simply screens out defects.
'Quality never happens by mistake, it is always the result of intelligent effort' (source unknown)
Zero defects
A step in the right direction has been the introduction of, so-called, 'zero-defect' programmes, particularly in the USA. Crosby believed that the traditional 'Acceptable Quality Levels' (AQLs), even if extremely stringent, compromised commitment towards zero defects.
The seeds of, perhaps, the most effective system for promoting quality in manufacturing, were sown just after WW2, during the reconstruction of industry in Japan. Americans, Juran and Deming, were instrumental in the subsequent success of Japanese manufacturing. The determination to succeed in that country, it has been suggested, explains the current difference between East and West in terms of quality and productivity. Japanese managers seized upon statistical and total factory quality management techniques, much quicker than their Western counterparts.
Total Quality Management (TQM)
TQM has been accused of being just another fad. However, it has evolved over many years in response to changing market demands. It will continue well into the future, albeit in some modified form under, perhaps, a different label. The basic concepts of TQM are very powerful, but also very simple. TQM puts the customer (the market needs) first, by imposing disciplines in all aspects of a manufacturing activity, right through from marketing, sales, product design, and manufacturing process to after-sales service. It provides the tools for feeding back information about the performance of all aspects of a business, thus making possible continual improvements, if the disciplines are rigidly imposed.
The Japanese approach to process improvement - a process is not just confined to a manufacturing activity - is to use simple, practical ideas (Kaizen). In the West it is probably considered to be smarter to use more complex solutions, with perhaps less predictable results.
Process improvement
Having examined the general principles of TQM, this review will now concentrate on process improvement, with emphasis on the semiconductor manufacturing environment. Firstly, it is necessary to define exactly what is meant by a 'process'. A process involves something acting upon something else. In a manufacturing process the main elements are:
- people
- equipment
- raw materials/piece parts
- methods/procedures
- environment (temperature, humidity, dust count).
These factors work together to produce an output ( Fig. 1). In an ideal world, process operator's performance is always 100%, machines never break down and raw materials (pieceparts) always behave in the same way. However, we live in the real world where things are a little different! Without some form of feedback mechanisms within the process, continual improvement would be virtually impossible to achieve.
Historically, in-process checks such as die shear strength, wire bond pull strength and package hermeticity, have been used on a sample basis to indicate the quality of the batch being worked on (the population). If the samples passed the set specification limits and the AQL was achieved, then the desired quality levels were deemed to be met.
With a little use of statistics, this data could provide a lot more information about the state of the process. The sample test results could form the basis of a system that might predict the future performance of the process ( Fig.2). This is the cornerstone of statistical process control (SPC).
Statistical Process Control (SPC)
SPC, if applied correctly, can be a very powerful tool for providing the feedback mechanism for continual process improvement. Two types of data can be used to monitor a process or a product. The first is variables data which is obtained by measuring some aspects of the process, for example, die shear strength or bond pull strength (the measurement must have units). If it is not possible to make practical measurements, then attributes data can be used. Examples of this, in microelectronic production terms, would be checking cropped package lead lengths on a go/no-go gauge or in a gross leak bubble test, where a steady stream of bubbles may be seen emanating from the defective package in the hot fluorocarbon liquid.
Both variables and attributes data can be used to plot control charts where, for example, average values (X bar) of the sample measurements (in the case of variables data) can be plotted against time, together with the range (R) of values in the sample (maximum value - minimum value). Mathematically derived upper and lower control limits complete the classic X bar + R chart ( Fig.3). The pattern of the plotted data over a period of time is then used to indicate whether the process is in, or out of statistical control. Data points outside the control limits indicate an out of control process, as do a series of points indicating an upward or downward trend in the results. Bunched results above or below the process average value, X double bar, also indicate that something is amiss with the process and rejects may be produced where the measured value is outside the specification limit for the product.
In addition to indicating whether a process is in, or out of, statistical control or, perhaps, about to go out of control, these data can be further used to determine how capable a process may be. Process capability (Cp), is the difference between the measured sample value and the specification limit value, usually set by the customer.
In mathematical terms:
where: total specified tolerance = upper specification limit - lower specification limit, σ = standard deviation of sample values.
In practice the Cpk value is more often used. This compares the spread of results between the upper specification limit and the process average (X double bar, or the lower specification limit and X double bar) and the actual spread of values from the process average (3 σ).
A Cpk value of 1.33 is usually taken as indicating an acceptable process capability, whereas a value of 1 or less is unacceptable. Process improvement can be indicated by an increasing Cpk value. Similarly, the capability of a new machine, for example a newly installed automatic wirebonder, can be measured in terms of the Cmk index (machine capability index). Measurements of bond pull strength, or ball bond shear strength are used to enable this value to be calculated. The machine can be said to be fully commissioned when a Cmk value of 1.33, or more, is regularly achieved. To obtain meaningful data, however, the materials used on the machine need to be constant in type and quality, to prevent unwanted variation in the process.
Pareto analysis
Pareto is a useful addition to SPC in improving the performance of a process, in that the frequency of defects found during the sampling process can be categorised. By putting the reject categories in rank order, the 'vital few' problems in a process can be separated out from the 'trivial man '. It is thus possible to focus the process engineering resource in the correct areas, for maximum effect.
Cause and effect
Isolation of errors in the process can be assisted by using the principle of cause and effect. Ishikawa (1976) in Japan developed the 'fishbone diagram' as a result of work within quality circles, aimed at identifying the relationships between cause of errors and effects of errors. The diagrams (see Fig.4) show the main variables in a process within a given environment.
These are expressed in terms of the 4Ms:
- Manpower
- Machines
- Methods
- Materials
When a process problem has been identified, perhaps from a previous Pareto analysis, a brainstorming session is held to list the possible causes of error within the process.
The 'fishbone diagram' is used as the framework for the generation of ideas. Further input can be contributed after the meeting and, frequently, the diagrams are displayed in the workplace to stimulate further thought. Each idea is evaluated, in turn, to determine whether or not it can be a major contributor to process errors.
Summary
The traditional approach to quality in manufacturing in microelectronics in the West, cushioned to some extent by a large proportion of military product, has been effective in screening out defective components. This has been achieved by sampling products at various stages of manufacture, e.g. die attach shear strength, wirebond pull strength, hermeticity and various visual inspection stages to very tight AQLs. Electrical testing and burn-in are accompanied by a comprehensive environmental testing regime to provide assurance that the proportion of defects within a lot is sufficiently low. Products manufactured under these circumstances are unlikely to be cost-effective and will almost certainly contain some proportion of defectives, however small.
An alternative approach has been used to great effect in Japan where the demand for high volume consumer products at low prices has grown steadily since WW2. TQM has evolved to meet these needs and encompasses a wide range of tools to provide active quality improvements. Most of the techniques are based on the analysis of data or information from a process, and feeding back corrective actions to enhance the operation and thus the product. The benefits of some of these methods, i.e. SPC, Pareto Analysis and Cause and Effect (Ishikawa), have been discussed. Changing world markets are dictating a move towards non-military products in the West. If this is to be achieved successfully, a parallel change in Quality Assurance policy, to a TQM orientated approach will be necessary.
