Making R&D lean: the logic of lean manufacturing has many possible applications in...

Large complex semiconductors are best optimized using feedback obtained from test wafers. However, running a single test wafer through a $1 billion semiconductor fabrication facility is expensive. In traditional chip development, such test wafers were done on demand using a "push" system. Chip designers would wait until they had enough issues to make it worthwhile to incur the cost of a test wafer. Then, test wafers would be scheduled, run and tested. It could take a month or more to

Today, companies like Taiwan Semiconductor use a radically different system, which resembles the "pull" systems used in lean manufacturing. Generic test wafers are scheduled to be run at fixed intervals. Multiple projects share the space on these wafers, and share their cost. Any project can get high-risk concepts into silicon quickly and get feedback in a matter of days. The shared test wafer is frequent and highly predictable; it acts as a window of capacity that "pulls" technical issues from the project teams into the manufacturing process. Since each project now bears only a portion of the fixed cost, it can use more test wafers, more frequently, and more cheaply than when it waited to request its own private wafer. With quicker answers, less effort is spent going down fruitless paths. This reduced downside enables more aggressive risk-taking. Rework costs drop; efficiency, quality and innovation rise. The use of "pull" causes simultaneous improvements in speed, cost and quality.

In this way, whether we recognize it or not, the compelling logic of lean principles is infiltrating R&D (see "The Logic of Lean," next page). These principles make enormous sense when they are applied properly, but they must be applied in a very different way in R&D than in the manufacturing domain where they originated. This article will discuss why lean principles are relevant to R&D and how these ideas can be applied safely.

Lean R&D is Not Lean Manufacturing

Although there are similarities between manufacturing and R&D, the differences are substantial. Manufacturing is a repetitive, sequential, bounded activity that produces physical objects. In manufacturing, risk-taking is not a major mechanism for adding value. A manufacturing process can do exactly the same thing a million times and still add value every time.

R&D, in contrast, is a non-repetitive, non-sequential, unbounded activity that produces information. Rational risk-taking is central to adding value in research and development. Unlike manufacturing, an R&D process adds no value when it does exactly the same thing twice.

These differences govern how lean principles should be applied in R&D. First, repetitive processes are very different from non-repetitive ones. In repetitive processes, all variability is bad and any reduction in variability will improve the process. In R&D, we cannot eliminate all variability without eliminating all value-added. We must take a more complex view of variability, carefully discriminating between "good" and "bad" variability. As we explain later, "good" variability creates economic value in R&D.

Second, manufacturing adds value to physical objects, and such objects can only be in one place at a time. This leads to an inherently sequential process. In contrast, R&D adds value to information, which can be in more than one place at a time. This creates opportunities to work non-sequentially, which in turn creates opportunities to use feedback in ways that are not available in sequential processes.

Third, R&D is not rigorously bounded in the same way as manufacturing is. In manufacturing we have fixed requirements: a defined start point and a non-negotiable finish line. In manufacturing, our work product must conform to these requirements; in R&D, we must constantly decide exactly what is good enough--we must continuously assess whether the economic benefit of further improvement justifies its cost. Emerging facts about the economics of our technology may cause us to stop short of our original goal, or to pass it because of unexpected progress. Thus, in R&D we must constantly adapt to emerging information.

Finally, taking rational risks is crucial in R&D. If we try to eliminate any possibility of failure, we forgo the chance to explore any promising new technology with uncertain prospects. Yet, such new technologies can create benefits, even when adjusted for risk, that far outweigh their costs. The inevitable result of investing in unproven technology is uncertainty: uncertainty in performance, uncertainty in cost and uncertainty in timing.

But We Still Can Use Lean Principles

These differences have a profound effect on how we apply lean principles in R&D. Certain approaches that produce benefits in lean manufacturing actually do harm in R&D. Other approaches can produce even greater benefits in R&D than they did in lean manufacturing. These increased benefits arise from two sources. First, variability is inherently, and necessarily, higher in R&D than manufacturing; lean principles permit us to adapt to this variability. Second, the acceleration of flow and compression of cycle time is much more valuable in R&D than it is in manufacturing. Lean principles enable us to achieve these two goals.

Now, let's consider ten important lean principles and how they can be adapted to the world of R&D.

1. Reduce Batch Sizes

In lean manufacturing, we reduce physical batch size. In R&D, our work product is information, so we reduce the batch size with which we handle information. For example, the discovery organization of one pharmaceutical company released its leads annually in one large batch. This meant that potential leads identified early in the year sat idle, waiting for the annual release date. A lean approach would release these leads in smaller quarterly batches. This reduction in batch size would cut idle time by a factor of four.

In R&D we can also reduce the "batch size" with which we explore the limits of a technology. By taking small steps into the unknown, we get fast feedback, enabling us to quickly truncate unproductive paths. For example, in software development there is a growing trend toward "agile methods" in which test feedback is provided to developers less than 24 hours after they write their code. With such rapid feedback, incorrect assumptions are discovered before they can become a defective foundation for subsequent work.

Small steps are particularly useful in R&D when the terrain we are exploring is shrouded in uncertainty. When we change many things at once, we introduce confounding factors that hinder our ability to diagnose the causes of success and failure.

Yet, many R&D processes are inherently structured to take large steps. One industrial organization commented that, "We do big projects because it takes as much effort to obtain a large bag of money as a small one." Whenever it takes a substantial effort to initiate a project, organizations respond by proposing large projects. But large projects have higher stakes and greater risks, due to their longer time horizons; therefore, they require more documentation and more thorough screening, further raising the effort to initiate them.

Viewed another way, the large batch approach tries to reduce risk by using meticulous forecasting to reduce the probability of failure. In contrast, the small batch size reduces risk by decreasing the consequences of failure using smaller, faster steps to get rapid feedback. Simultaneously, it reduces the probability of failure by shortening planning horizons, which reduces forecasting errors. Some companies implement this small batch approach with "piecewise funding"; additional funding only becomes available when key milestones are met successfully.

Finally, large batch sizes also have insidious psychological effects. Organizations constantly have trouble disengaging from large expensive programs. Psychologically, people interpret ambiguous signs of success more positively in proportion to the amount of effort they have invested. Furthermore, the delayed feedback associated with long programs removes the sense of urgency found in shorter ones.

2. Make the Process Tolerate Necessary Variability

In lean manufacturing, variability is always bad and should be eliminated; in R&D, life is not this simple. We need to carefully distinguish between good and bad variability. We cannot add value without trying something new, and we cannot try something new without introducing uncertainty and variability into our outcomes. This necessary variability, which is the natural partner of rational risk-taking, is fundamental to adding value in R&D.

What complicates things in R&D is that not all variability is necessary. It is very desirable to eliminate the variability that comes from sloppiness, inattention, lack of foresight, and endlessly repeating the same mistake. Our challenge in R&D is to distinguish between the two types of variability, eliminating one and exploiting the other.

In lean R&D, instead of eliminating variability, we try to capture the value of "good" variability, while minimizing its cost. We can do this by structuring and operating our processes so that they function well in the presence of variability. Consider a few examples:

First, we need excess capacity for variable activities on the critical path. If we operate processes having variation at high levels of utilization, this amplifies variability within the process (1). For example, consider a testing area loaded to 100 percent utilization. At this operating point, a small variation in arrival time can send a job to the end of a large queue. Missing arrival time by one day may result in a three-week slippage for the project. Often, a small increase in capacity (e.g., 10 to 15 percent) can eliminate most of these slips.

Second, we can control key process queues. For example, one organization monitors queue size in its engineering change process and adds resources whenever the queue gets too large. By actively controlling queue size it dramatically reduces the variability of cycle time through the process.

Third, we can invest in creating flexible resources. The culture of R&D commonly values narrow specialization. Our experts are rewarded for their depth in a narrow area, which creates bottlenecks. In contrast, in lean factories it is common to see 10 to 15 percent of workers cross-trained to work in any area. This provides the flexibility to move resources to emergent queues. We will return to this issue later.

Finally, we can avoid using large batch sizes. Large batches inherently add variance to the workstream. Anyone who has exited a full jumbo jet, and entered a customs area, can appreciate that if a large spike in demand is not matched by an equal spike in capacity, a queue will result.

3. Focus on Maintaining Flow Instead of Perfect Planning

Planning relies on our ability to forecast the future. It is a powerful tool when we can forecast the future. A useful metaphor is walking down a road in the fog. Using the perfect planning approach, we would wait at the beginning of the road until we could see the entire path. When we focus on flow, we begin walking over the short distance that we can already see. Then, as we progress, more of the path becomes visible. Although we only plan as far as we can see, by moving forward we ultimately can see everything.

In lean manufacturing we maintain flow by creating mechanisms to dynamically adjust capacity and demand. We pull resources to the emerging queues and pull demand to available capacity. In R&D we have the same potential. By carefully monitoring flow rates and queue size, we can make adjustments to maintain flow. Importantly, this requires dynamically responding to emerging information, not preplanning everything far in advance.

In a certain sense, this adaptive approach simply recognizes that technical problems arrive at random times and in random sizes. If 100 percent of our resources are committed in one large batch at the beginning of the R&D year, we have fundamentally mismatched the arrival process of work and the arrival process of resources. This will inevitably lead to delays and inefficiencies, as we try to disengage resources to deal with emergent problems.

4. Pull Don't Push

Most conventional R&D planning systems rely on "push." Corporate strategies drive product strategies; product roadmaps drive technology roadmaps; technology roadmaps drive project plans; project plans drive monthly resource allocations. Everything is carefully scheduled in advance, as if it were predictable. But R&D is not predictable.

Pull-based systems take a different approach. Daily assignments of goals and resources are made in response to the current status of work. Resources may be pulled to meet variations in demand, or demand may be modulated to fit the availability of resources. The speed of these adjustments reduces variation, and reduced variation leads to smaller average queues, smoother flow and faster cycle time.

Consider the approach of one software organization: some programmers are cross-trained to do testing; when the queue of untested code reaches a specific size, these programmers are shifted from programming tasks to testing tasks. By "pulling" this resource, the company reduces queues, shortens cycle time, accelerates useful feedback, and smoothes flow.

5. Create Fast, Powerful, Feedback Loops

Active control systems permit airplanes with unstable aerodynamics to fly smoothly. Similarly, R&D processes can operate in more turbulent environments when they incorporate fast, powerful feedback loops. The small batch sizes discussed in Principle 1 are necessary for fast feedback, but they are not sufficient. We must actively use information from a downstream step to make changes in an upstream step. For example, in our test wafer example, data from early test wafers modifies our design approach for circuitry that is not yet designed.

Interestingly, fast feedback also produces important psychological benefits. Researchers feel more in control, are more willing to take risks, and are more inclined to use initiative, when unproductive paths can be quickly truncated.

6. Requirements Are Seldom Required

As mentioned earlier, in manufacturing failing to meet requirements is not an option. In R&D it is one of our most powerful tools for managing flow. In R&D, requirements exist to signal which attributes of a technology are most highly valued. Projects attempt to excel in these attributes; however, they should dynamically react to emerging information regarding the cost of improving these attributes. For example, a consumer product company was developing an energy technology. As experimental data emerged, it became clear that meeting the original requirements would require achieving more than 100 percent thermal efficiency. At this point, because meeting requirements required breaking the laws of physics, the project goals were modified. In R&D we must be willing and able to modify our goals in the presence of compelling new information. This is a crucial difference between lean manufacturing and lean R&D.

7. Invest in Flexibility

Lean manufacturing maintains flow by making resources flexible. Capital equipment is selected on the basis of flexibility; production workers are cross-trained to do the jobs of other workers; work is organized into cells where workers can help co-workers; production lines are laid out to enable workers to easily assist on other workstations. Even highly repetitive environments like manufacturing have enough variation to make this flexibility worthwhile.

In R&D, variation is much higher and the benefits of flexibility are much greater. However, because roles are rigidly circumscribed in most R&D organizations, it is harder to create flexibility. Yet, it is still possible. In many research-based start-up companies, scientists routinely work on whichever task is currently delaying progress, even when it is outside of their primary area of expertise.

Flexibility is a matter of both attitude and preparation. The attitude is created by incentive systems and recognition for individuals who work outside of specialty. Preparation involves broadening skill sets before they are needed. One very successful telecommunications R&D organization uses its annual bonus system to reward staff members who significantly increase their technical breadth during the year. In contrast, most of its competitors reward technical depth, thereby creating inflexible staffs of narrow specialists.

It is important to recognize that we don't need to create completely interchangeable people to get flexibility. Any ability to remove tasks from the worklist of a bottlenecked expert helps maintain flow.

8. Achieve Adequate Failure Rates

The results of any experiment can be viewed as a communication process between the physical universe and a researcher. Experimenters receive messages that either confirm or refute their hypotheses. Viewed through the perspective of information theory, an experiment generates information most efficiently when its probability of success is 50 percent (2). If experiments have either too low or too high a probability of success, it means they are using resources inefficiently and creating waste. "Efficient" failure rates create less waste than trying to "do it right the first time."

Lean manufacturing manages risk by trying to make the probability of failure zero. Lean R&D manages risk by accepting higher, "efficient" failure rates and limiting the downside consequences of these failures. As mentioned earlier, this is done using small batch sizes and fast feedback.

9. Understand the Economics of Waste

A chronic mistake when using lean principles in R&D is to seek waste in the same place that it was found in the factory. In manufacturing, expense budgets are high and cycle time is measured in days or weeks. In R&D, on the other hand, expenses are low and cycle time is measured in months or years. As Figure 2 illustrates, the relative economic importance of wasted cycle time, compared to wasted expenses, can be over 200 times greater in R&D than manufacturing.

Lean manufacturers identify their waste by paying careful attention to economics. It is just as important to do this in lean R&D, because the economic importance of different forms of waste can vary by orders of magnitude. For example, at chemical and pharmaceutical companies it is common to see heavily utilized pilot facilities. This would appear efficient, but a simple analysis would show that the delays caused by such heavy utilization are far more expensive than the cost of an underutilized facility. When we ground the implementation of lean R&D on economics, we can make much better tradeoffs between objectives.

10. Control the Right Parameter

Understanding economics is also the key to controlling economics. Prior to lean manufacturing, United States manufacturers focused on efficiency as the key measure of performance. This led to specialized equipment, loaded to high levels of utilization, using large production batches, and run by specialized workers. Instead of leading to superior profits, this approach produced enormous waste in the form of inventory and overhead.

Today, some R&D organizations assume that lean principles primarily involve the improvement of efficiency by eliminating wasteful expenditures. Yet this may actually be the least useful aspect of these principles. For example, some R&D groups measure the efficiency of support groups and try to maximize it. But, efficiency is maximized at high levels of utilization, and high levels of utilization in the presence of variability produces service delays. We have seen such high levels of utilization force highly-paid technical employees to perform their own support tasks, just to avoid waiting for support groups. By locally optimizing the efficiency of the support group, severe waste is created at the level of the overall process.

Consider, for example, the tooling shop at one manufacturer of medical devices. When it was measured on tooling cost per project, tooling was cheap but it was always getting on the critical path. Then the company recognized that the cost of delay was far more important than tooling cost. As a result, the relevant metric became how often tooling got on the critical path of projects. This led the shop to put in extra capacity, begin using overtime, and use outside resources to eliminate delays. Reduced delay costs far exceeded any additional tooling costs.

Summing Up

To blindly transfer the concepts of lean manufacturing into an R&D environment may do more harm than good. R&D adds value in a very different way. Behaviors like rational risk taking are central to success in R&D, but add little value in manufacturing. Variability is pure waste in manufacturing, but can be the essence of value-added in R&D. Nonetheless, if we respect the unique nature of R&D, we can find many applications for the logic of lean. Fortunately, the best R&D managers recognize the important difference between adapting underlying principles and blindly copying superficial behaviors.

The Logic of Lean

It is important to recognize that what is called "lean manufacturing" is actually a system created by combining the logic of lean principles with the economics of manufacturing. When this system is optimized, we call it, "lean manufacturing." When the same lean principles are combined with the economics of R&D they result in a very different system. When this system is optimized, we call it, "lean R&D."

First, let's review how lean principles work. About 50-60 years ago, Japanese industrial engineers like Taiichi Ohno (3) and Shigeo Shingo (4) developed an approach to manufacturing that is known as the Toyota Production System (5). The Western world calls this approach, "lean manufacturing." As Ohno explains, "The basis of the Toyota production system is the absolute elimination of waste" (3, p. 4). As stated, this provides little insight. Western manufacturers had focused on eliminating waste since the beginning of the Industrial Revolution; however, they focused on direct labor, direct materials and utilization of capital assets. What was different about the Japanese approach was that it expanded the focus to wasted time and overhead.

A key method for reducing waste is an approach called just-in-time (JIT) production. JIT dramatically reduces batch size, which results in proportional reductions in both inventory and cycle time. Traditional manufacturing doctrine predicted that such small batch sizes would result in great inefficiency, because the fixed cost of producing each batch would be incurred too often. However, the Japanese experience with small batch size defied this prediction. Smaller batch sizes produced simultaneous improvement in speed, cost and quality. The improvements in speed were expected, but the great improvements in cost and quality were not. Quality improved because smaller levels of inventory caused downstream processes to provide quick feedback to upstream processes. The old, large-batch process might deliver 10,000 parts only to discover it began producing bad parts partway through this large batch. When batch size dropped to 10 parts, problems with an upstream process produced almost instant feedback from the downstream process. There were also reductions in overhead costs. Shorter cycle times led to fewer jobs in process, which led to less overhead to schedule, track and control the individual jobs.

A second key element of JIT is the creation of flow. Flow occurs when value is being added continuously--work products never sit idle and motionless. This flow is created by using control systems based on "pull" rather than "push." This terminology originated when Taiichi Ohno observed how American supermarkets worked. Customers "pulled" product from the shelves. When stock on the shelf got low, it was replenished by "pulling" product from the stockroom. In contrast, a "push" system would try to forecast exactly what each customer would buy, and exactly when they would buy it. It would schedule deliveries based on these forecasts, regardless of whether the shelves were full or empty. As accurate forecasting would be almost impossible, such a system would result in shelves overflowing with products that were not selling and out-of-stock for products that were selling.

Pull-based systems represent a simple but sophisticated way to adapt to varying demand. Such systems smooth flow, reducing both queues and cycle time. The interacting elements of lean are depicted in Figure 1.

[FIGURE 1 OMITTED]

With care, these two central elements of lean thinking can also be applied in R&D. Care is required because the work process, economics and mechanisms for adding value differ substantially between manufacturing and R&D.--D.R. and L.S.

References and Notes

(1.) Reinertsen, Donald G. Managing the Design Factory. New York, NY. The Free Press, 1997. The toxic effects of high-capacity utilization are discussed in more detail in Chapter 3.

(2.) The need for 50 percent failure rate is explained in Chapter 4 of Ref. 1.

(3.) Ohno, Taiichi. Toyota Production System: Beyond Large Scale Production. New York, NY: Productivity Press, 1988.

(4.) Shingo, Shigeo. A Study of the Toyota Production System from an Industrial Engineering Viewpoint. Revised Edition. Portland, Oregon: Productivity Press, 1989.

(5.) Monden, Yasuhiro. Toyota Production System: An Integrated Approach to Just-in-Time. Atlanta, GA: Engineering and Management Press, Institute of Industrial Engineers, 1998.

Donald Reinertsen is president of Reinertsen & Associates, in Redondo Beach California. He has specialized in improving product development processes for the last 25 years. He is co-author, with Preston G. Smith, of Developing Products in Half the Time (John Wiley, 1998), and author of Managing the Design Factory (Free Press, 1998). He teaches an executive course on product development at the California Institute of Technology, and has an M.B.A. from Harvard Business School. don@ReinertsenAssociates.com

Leland Shaeffer is a principal at Reinertsen & Associates. He has worked in engineering and marketing positions in technology-intensive businesses for over 20 years. He co-teaches with Reinertsen at Caltech, and has an M.B.A. from Stanford University. lee@ReinertsenAssociates.com