Heres the case for back-end factory automation
11/01/1998
Here`s the case for back-end factory automation
Mitchell Weiss
Careful analysis of semiconductor manufacturing`s "back end"- test, assembly, and packaging - shows that investment in factory automation is significantly out of sync with investment in automation for wafer processing or the "front end." I`m talking about "automation" that is a little outside of the traditional numbers tracked in the semiconductor industry. It involves "materials logistics" - the transport, delivery, loading and unloading, and control of materials in the back end.
One of the mistakes you can make when discussing the back end is to say that there is no automation. (I`m not talking about automation of individual process tools; indeed, wire bonders are one of the coolest pieces of automation on the face of the earth. Automatic IC handlers are used extensively, but not in the context of "factory automation" that we see in the front end.) Factory automation components include:
the automated material handling system (AMHS),
the material control system (MCS),
tool-loading systems, robotic devices,
manufacturing execution systems (MES),
cell controllers, and
dispatch and scheduling systems.
Consider that almost all 200-mm wafer fabs use some type of AMHS. In 1997, semiconductor manufacturers spent roughly $500 million on the front-end AMHS and control, but virtually nothing on back-end AMHS.
Interestingly, the drivers for back-end and front-end AMHS are the same. For example, cleanliness is increasingly important; the industry is seeing more back ends built to Class 1000 or better standards. Proportionally, shrinking geometries are as much a part of the back end as the front end; today, the leading edge is 70-?m pad-to-pad dimensions.
One distinct difference between the back end and the front end is that the back end has a variety of material forms - die, magazines and carriers, and packages - while the front end has one single wafer size. In addition, the lifetime for this wide variety of material forms is dramatically short compared to wafer sizes, driven largely by chip-scale packaging trends.
Traditionally, the word has been that the back end is not as tolerant as the front end to the high cost of tools. I`ve heard it said, "You can`t sell that way to the back end; it is not as rich as the front end." But the fact is that back-end spending comprises about 30-35 percent of the capital spending in this industry, according to market numbers from VLSI Research. That`s a lot of money. While a wire bonder may only cost $100,000 or a die bonder $250,000 (both relatively cheap compared to a multimillion dollar stepper), there are many more wire and die bonders used in the back end compared to steppers in the front end. And don`t forget that automatic test tools are outrageously expensive.
Keeping those expensive testers and all other back-end systems fully utilized is just as important as keeping a stepper utilized on the front end. Recently, the industry has become accustomed to looking at each individual tool`s overall equipment effectiveness metric. But isn`t the real task - and the most important thing - keeping a given tool set utilized?
You need to look at the whole factory. If it is only running at 60 percent capacity, you are leaving 40 percent of your investment money on the table. This, in part, is the role of AMHS and control. Factory automation can provide increased equipment utilization, improved factory balance, execution of more complex capacity plans, improved material control and tracking, and enhanced safety and ergonomics.
Addressing back-end material handling has been somewhat difficult because there aren`t any good cost models. We have begun to develop them at PRI Automation for just the assembly side of the back end, looking at all costs involved in assembly, material, equipment, labor, space, energy, etc., ending with a cost/station metric. The first role of this modeling is to answer the question: "For every tool in an assembly area, how much money do I have to add to the price of a tool to accept automated material handling?"
This modeling has shown that even with just a 5 percent utilization improvement at a given tool for a fab`s different package scenarios (SOICs, MPUs, and DRAMs), the payback for investment in factory automation can occur in just eight months. For the operations manager, this means that given a pretty good automation budget, payback is easily justified through just small utilization improvements. Importantly, we determined this by just tweaking utilization sensitivity in our modeling. Factoring in labor savings and other elements could yield an even shorter payback.
Okay, you ask: Since there is proven technology for the front end, why hasn`t automation happened in the back end? The answer is that front-end technology isn`t ideally suited for the back end for a number of reasons, including the greater variety of material forms and the absence of standard carriers. The back end also has a greater number of tools, which have to be accessible for changeover and are often, importantly, legacy tools. The lifetime of a wire bonder is much longer than the lifetime of a stepper. One of the tricks of any back-end factory automation problem becomes retrofitting into these legacy environments.
Thus, it`s a little trickier despite the incentives. We just have to start the ball rolling. Overcoming these challenges and performing the required development is only a matter of time; the potential seen in disparate investments in front-end and back-end factory automation is destined to change. For example, according to VLSI Research data for 1997, ~$25 billion was spent on semiconductor manufacturing equipment, ~$1.5 billion on automation, and ~$0.1 billion on test, assembly, and packaging automation. The latter was principally for wire bonder and die bonder cells, such as those from ESEC and Kulicke & Soffa, and includes both software and hardware. This means that the industry spent significantly more for front-end factory automation than for back-end automation.
From the available data, we see that test, assembly, and packaging are about 30 percent of the spending for equipment, but that back-end automation spending is clearly <10 percent of the automation spending. In the front end, spending for automation is about 2-3 percent of total front-end spending. (Interestingly, this latter percentage should grow in the near future.)
To equate spending for factory automation in both the front and back end, the industry needs to spend roughly 3 percent of the money going into back-end equipment for back-end factory automation, rather than <1 percent today. In this scenario, if equipment sales reach $37 billion by 2000, $12 billion of that should go into equipment for assembly, test, and packaging, and $3 billion into back-end factory automation.
When putting semiconductor manufacturing`s back-end spending in context with front-end spending, you can see that the back end isn`t the cheaper stepchild that it has been dubbed. You`ll be amazed to discover how far out of line spending on factory automation has been.
For FREE info circle 16
Mitchell Weiss is VP of strategy and technology at PRI Automation Inc., 805 Middlesex Turnpike, Billerica, MA; ph 978/663-8555, fax 978/663-9755.