Semiconductor assembly practice

The clean-room environment

The American Federal Standard 209B Clean room and work station requirements - controlled environment uses categories of clean rooms such as ‘Class 10,000’, where the figure is the maximum number of particles of 0.5µm or larger in each cubic foot of the environment. As shown in Table 1, the less stringent standards also allow a smaller number of larger particles.

Table 1: Maximum particulate counts permitted in a cubic foot of air
(US Federal Standard 209B)
  Size (µm)
Class >0.5 >1 >5
100,000 100,000 20,000 700
10,000 10,000 2,000 60
1,000 1,000 200 0
100 100 20 0

The classes from the American Standard have entered the vernacular, but European equivalents such as BS5925 are similarly defined, although in metric terms (Class 100 equating to 3.5 particles per litre). Room cleanliness is defined as the average particulate contamination through the entire room, as determined by sampling techniques.

The numbers of particles permitted seem very high, but have to be viewed in the context of the number of particles that people generate, some typical values for which are given in Table 2.

Table 2: Particle emission by humans
particles >0.5µm/minute activity
100,000 Standing or sitting still
500,000 Sitting, with slight head, hand or lower arm movement
1,000,000 Sitting, with moderate body and arm movement and slight foot movement
2,000,000 Standing, with full body movement
5,000,000 Walking slowly
7,500,000 Walking normally
10,000,000 Walking briskly
15–30,000,000 Exercise and games

In conventional (‘random flow’) clean rooms, air is supplied at ceiling level and returned through low level grills. The recirculation system uses HEPA filtration and controls temperature and humidity as well as reducing dust count. The number of air changes per hour rises exponentially with the standard of cleanliness being sought, and can be very much higher than would be required just to control temperature and humidity. Careful design of the ventilation system is needed to ensure that the clean room is economically viable to run, bearing in mind the energy cost and the costs of filtration equipment maintenance.

To achieve the standards of cleanliness required for semiconductor assembly, it is therefore not uncommon to use a combination of a moderately clean room (Class 10,000) and laminar flow benches, where locally clean conditions are provided in the areas where critical operations are carried out.

In the laminar flow bench, clean air is introduced through a full HEPA filter wall, generally at the back. The level of cleanliness achieved is a function of the air flow and the distance from the filter wall.

Clean rooms are sealed, with a small positive ‘over-pressure’ to reduce the amount of contamination that seeps in from outside, and special attention is given to reducing the adverse impact of product and people entering and leaving the room. Materials are often passed through a double hatch system, and personnel go through an air lock, with tacky floor mats to trap dirt carried on shoes and trolley wheels: entry should be restricted to a minimum number of essential personnel.

Working areas are kept clean partly by providing the right airflow, and partly by reducing particulates at source:

All these are elements of cost which a semiconductor assembler has to consider.

[ back to top ]

The importance of controlling contamination

Assemblies can be contaminated during processing in a wide variety of ways, by the equipment, by the environment, and especially by the operators. One author described the clean room worker as a ‘walking torch of contamination’. This contamination contains organic matter, such as oils, as well as sodium, magnesium, aluminium, silicon, phosphorus, sulphur, chlorine, potassium, calcium and iron. Cosmetics contain extra organic materials that are difficult to remove, as well as bismuth, barium and titanium.

Contamination usually remains dormant until activated by moisture, acting ‘like a time bomb with an unknown length of fuse’. For example, work carried out in 1984/5 following the failure of a space shuttle found that dried human spittle plus moisture had corroded the aluminium metallisation on a die, resulting in expensive catastrophic failure.

Back-lit video recordings made of a person talking showed that certain initial consonants (f, p or t) projected small amounts of saliva for up to a metre, whilst coughing could contaminate with saliva and dead lung tissue up to two metres away, and sneezing projected saliva for between three and five metres at speeds as high as 350km/hr! Contamination could be found in the product after any operation where the operator was not physically isolated from the chip. Surprisingly, it was found that some automated machines allowed much longer periods of direct exposure than older manual bonders.

In another case, ion implant equipment was found to be charging the wafer, causing it to attract all particles which came loose during the operation of the mechanical loader. This indicates how the design of the equipment and process can determine whether or not the combination of operator, machine and operation will contaminate the part.

Changes in equipment design and processes resulting from these experiences have been accompanied by a trend for operators to wear partial face masks to reduce the incidence of spittle problems. For improved reliability, the preferred solution is to use closed cassette-to-cassette transfer, where the die is never exposed to the environment. But trays and handling jigs have to be cleaned, and the point has been correctly made that ‘parts can seldom be cleaner than the trays that contact them’.

Dealing with contamination

Particle contamination , consisting of hair, skin particles, fibres, lint, metal chips, abrasives and other small particles, may be picked up at all stages of production. Cleaning procedures such as agitation in water will remove these, but prevention is better than cure.

Organic contamination may be removed by solvents, but a more normal method is to use an RF-excited plasma to perform low temperature ‘ashing’.

Ionic contamination from water soluble acids, plating residues and other metal salts will reduce the operational life of a circuit and can be removed only progressively by washing repeatedly in pure water.

Sources of contamination can be traced, by using polarising microscopes and other means of identifying both the material and the size of particles.

[ back to top ]

Clean-room materials

Traditional paper is made of wood pulp and produces substantial quantities of dust in use. General good practice is therefore to eliminate this by employing paperless methods for record keeping. Unfortunately, having some printed materials in the clean room is almost inevitable.

The first clean room paper substitute to be developed was made of Tyvek®, a high density polythene from DuPont. It is durable, relatively clean, free of cellulose, does not tear, and can be written on with ball-point pen and printed using an impact printer. However, it will not tolerate the high temperatures of laser printing or photocopying. An alternative which will withstand these processes is made from wood pulp saturated with latex, which generates 90% fewer particles than ordinary paper. Such papers now supply the majority of clean room stationery needs for Class 100 or below.

A third synthetic material used for clean room stationery is silica-filled polyethylene. This micro-porous synthetic substrate has the advantages of Tyvek®, combined with good heat resistance and absorption.

One has of course to be careful with other sources of contamination:

Choosing and using clean-room wipes

To maintain the effectiveness of clean rooms, careful cleaning of walls, floors and work surfaces is needed, and wipes are used both wet and dry for this. Wipes can be made of polyester, polypropylene, nylon, cotton and cellulose, materials which are either fibrous or made into fibres. Synthetic materials are generally made into monofilaments (unbroken single strands) and natural materials made into staple (short) fibres. Staple fibres are twisted into yarns to make a knitted or woven fabric, or put into a slurry and the liquid pressed out, making a felt.

Fabrics made from monofilament fibres are inherently less contaminating than those made from staple fibres. However, wipers have to be cut from fabric, a process which can release fibres and loops of filament. To remove these contaminants for the cleanest applications, the edges are sealed and the wiper laundered.

Wipers have to be effective in picking up wet and dry contaminants and holding them for disposal, without depositing significant amounts of contamination or being too costly. Depending on the situation, these considerations will have be given different emphasis.

In dealing with spills, a dry wiper with a substantial capacity for holding the spilled liquid is required, wiping slowly to maximise liquid uptake. In most other cases, however, a damp wiper is preferable, as damp fibres hold particles better than dry fibres. The liquid reduces the adhesion of the particles to the surface to be cleaned, and any films can first be dissolved, and then transferred to the body of the wiper as it absorbs the liquid.

As in any cleaning procedure, the cleanest wiper should be used at the end of the process, and on the part which has to end up cleanest. Surfaces should therefore be wiped systematically from cleanest to least clean, folding and unfolding the wiper frequently to create a fresh surface.

after Cooper 1995

[ back to top ]

Automation, yield and cost issues

In a manufacturing environment, cost and yield are paramount, and related to both the quality of process control and the degree of automation. Even at modest levels of output, individual machines will be highly automated, with the strip of lead-frame being the initial unit of manufacture and transport between the machines using cassettes to provide a safe environment for the product. The differences that might be expected for high volume manufacture would include using a matrix lead-frame instead of a strip and interconnecting the equipment rather than involving separate handling. In a typical set-up, the wafer saw has high capacity and would feed a number of lines made up of perhaps one die mounter, three wire bonders, and two ‘automatic’ mould machines. Wire bonding is often the limiting process for a line, but the balance depends on package size and lead count.

Costs are always given as an addition to the price of the chip, and will vary greatly according to volume, and package style. As a rule of thumb, for Hong Kong assembly, the cost of a plastic-packaged part in volume is 1¢/lead plus the die cost. And this includes the costs of running a facility, including consumables such as deionised water, gowning, masks, gloves, extraction and plating.

The breakdown of costs will vary enormously, reflecting the product type, the level of automation and company accounting policy:

labour 10–40%
materials 10–15%
factory costs and equipment depreciation 45–80%

Care has to be taken when calculating and comparing figures for yield. Whereas a test yield of 95% including device burn-in may initially seem quite impressive, the failure rate may have been depressed by die problems and mask a true assembly yield of over 99.5%. At the prototype stage, die failure rates at burn-in can be very high (up to 30%), which requires good feedback between assembler and wafer fab to eliminate failure causes, and a lot of effort put into probing, to ensure than only good devices are assembled.

For those used to the lower yields of soldering processes, such figures may seem high, but reject levels of 2–3 per thousand are needed in order for European assembly operations to compete with the lower costs of the Far East. One of those interviewed stressed that the company’s good results for productivity and yield did not reflect ‘black art’:


“For die bond it is a question of the dispenser and material choice; for wire bonding a combination of accurate equipment, which is programmed correctly, calibrated and characterised.”


[ back to top ]