Life Cycle Considerations

Carnegie Mellon University
18-849b Dependable Embedded Systems
Spring 1999

Author: Philip Koopman

Abstract:

The life cycle of an embedded system varies dramatically, from processors embedded in disposable consumer goods to applications requiring maintenance and support for decades. Designing an embedded system often requires taking into account the complete product life cycle, from initial product concept, through its operational period, and into replacement with newer equipment. While the design phase is covered by other topics, areas of specific concern to a life cycle perspective are: an accurate life cycle economic model to guide engineering tradeoffs, taking into account requirements for logistics and support over the product operational period, and issues specific to refurbishing/retiring/discarding the system at end-of-life. While the term "life cycle" has different meanings to different technical communities, the central idea is to expand the traditional engineering emphasis on the "design cycle" to encompass optimizing utility, profits, and tradeoffs across the entire lifetime of the embedded system being designed.

Introduction

The idea of a product life cycle acknowledges the fact that designing and selling a product is only part of the story. In fact, every product goes through a series of steps between the time it is first conceived and the time the manufactured product is retired or discarded. Figure 1 shows one view of the various phases of a produce lifecycle [ref to edrc material].

Figure 1. The phases of a product life cycle. (click for larger version)

The product starts with a need or opportunity in the marketplace. In embedded applications, this need has historically been a goal of reducing cost or increasing the functionality of an existing electromechanically controlled system by inserting computer technology. An example is digital engine controllers for automobiles, which were virtually required to meet tougher emission controls that could not be handled by mechanical engine controllers. At other times, completely new opportunities exist because of the progress of technology. As a somewhat frivolous example, greeting cards that play music or recorded messages require fairly sophisticated, inexpensive electronics for operation.

Once a need or opportunity is defined, a concept for a product is created. As an example, a product concept might be something like a digital cell phone instead of an analog one, in order to reduce audio noise heard by the user. With a life-cycle approach this stage involves not only defining the product, but also addressing the business model of how the company will make money from the product, how users will employ the product, how the manufacturer and/or user will benefit from any potential upgrades, and how the product will be retired, disposed of, or refurbished.

After concept development, both the product and manufacturing process are designed. In some cases an existing manufacturing process is taken as a fixed parameter, but often it is more efficient to permit changes to the product to accommodate manufacturing needs as well as changes in the manufacturing process to accommodate product needs. In many embedded systems the manufacturing volumes are quite high, making it worthwhile to modify or even create new manufacturing processes. Additionally, in smaller companies that create embedded products, much of the manufacturing process may actually be created via contract manufacturing, meaning that creating the manufacturing process consists largely of selecting vendors for various process steps and then arranging them in the appropriate order rather than arranging manufacturing equipment at the company's own plant.

Within product design, it is common enough to take the approach of designing something that functions correctly and can be manufactured for a profit. However, with a life cycle approach it is often better to maximize profit across the entire lifecycle, even if that means reaping less product on the initial equipment sale. For example, a system that is to be upgraded might be designed with higher-priced component sockets or connectors that support multiple installations and removals, rather than less expensive sockets that can only be used a single time. (There is, obviously, a tie-in to the business model at this point, and there may be structural impediments within a company that prevent such global optimization.)

Once the design process is completed, the equipment is produced and deployed. Considering for these life cycle phases in product design can reap significant profits. For example, automatic component placement machines can only handle a fixed number of different components, so designing to stay within that limit either speeds up production throughput or avoids the cost of buying a second placement machine. Similarly, considering deployment requirements during the design phase can be quite beneficial. For example, a user interface on a new piece of equipment might be made to mimic the user interface on an older, non-computerized piece of equipment to reduce the need for customer training even though a radically new interface might be slightly more efficient or cheaper to manufacture.

After the product is deployed, it must be supported and maintained. Support and maintenance for desktop software is dominated by technical support phone calls with problems of installation, bug reports, and difficulty using advanced features. Desktop hardware maintenance far less frequent, and involves situations such as disk crashes, dirt, and power supply failures. Embedded systems seem to have a somewhat different focus for support and maintenance. Since the embedded software is often more focussed on controlling some other system, the amount of software functionality exposed may be significantly less than in a software product such as a spreadsheet or database. However, user interfaces are often minimal, making it difficult for the user to access the available functionality (the classical example is setting the time on a VCR). Thus, technical support is more likely to be help due to user interface obstacles or mismatches between the consumer's mental model of operation and the actual software design. From a maintenance point of view, interactions with other products are less likely to be an issue since the embedded system comes with its own dedicated computer. However, embedded maintenance is complicated by the fact that problems may be due not only to a computer failure, but in fact are more likely to be due to failures in sensors, actuators, or mechanical components. Thus, in support and maintenance it is important to provide extensive diagnostic and monitoring in complex embedded systems to facilitate repair.

The need for upgrades varies dramatically depending on the life expectancy and usage of the embedded system. At the low end, some embedded systems are completely disposable, such as the electronic greeting card mentioned above, many toys, and consumer products in which repair labor is more expensive than the product, or in which products are evolving so quickly that products are obsolete before they have a chance to break (this is probably the case with digital cameras at the moment). On a larger time scale, however, enduring embedded systems must be upgraded once or more during their product life. For example, elevators can remain operation for up to a hundred years or more, but are upgraded every decade or so. The same is true of many expensive pieces of equipment such as military weapon systems or aircraft, which must be operated for many years to amortize purchase costs. With equipment that is likely to be upgraded, it is important to design the system in such a way that upgrades can be done cost-effectively, and that system downtime is minimized during upgrade. Perhaps the most extreme example of design-for-upgrade is in telephone switches, which are required to operate continuously through a system upgrade.

Finally, all embedded systems are eventually retired, discarded, or replaced. Designing the system to be retired gracefully can significantly reduce costs to the manufacturer, the user, or society as a whole, as is discussed in a later section.

Key Concepts

While the notion of a life cycle encompasses a very broad range of topics, there are three particular life cycle areas that are not covered in other topic discussions: life cycle costing, logistics, and design-for-disposal (also known as life cycle assessment, or "green design").

Life Cycle Cost

The life cycle cost of a product includes not simply the cost of materials and labor to manufacture it, but in fact all costs associated with the product from inception to retirement. The idea of a life cycle approach to cost is not specific to embedded systems, but rather is more generally applied to very expensive capital purchases such as buildings, factory machinery, and military systems (ships, planes, tanks, etc.). However, since embedded system designers are often embedding computers in such expensive systems, it behooves them to understand the financial model for life cycle costing so that they can take this into account in their work.

Kirk & Dell'Isola's book [Kirk95] provides a comprehensive look at life cycle costing from the perspective of operating a commercial building, such as an office building (the following discussion is based on material from that book with some augmentation). However, the concepts they discuss apply to many embedded systems in general. (As an aside, an office building is in fact an embedded system. There are computers controlling the climate, operating the elevator system, and in many cases controlling the lighting. In some buildings these three systems are beginning to be coordinated for efficient operation as well as lower cost maintenance.) In general, the factors included in the life cycle are those discussed in the previous section with respect to design and retirement of the product. In general, the idea is to minimize total cost for owning and operating an embedded system over the complete life of that system. This means that in addition to technical design factors, specific economic factors must be considered.

The life of a product is the shortest of three different aspects of system life:

Useful Life (utility). This is the obvious notion of equipment lifetime, in which eventually equipment wears out to the point it is beyond reasonable repair.
Technological Life (obsolescence). A system may become expensive or impractical to maintain even though it still is theoretically repairable or operable in general. For example, it may be impossible to find technicians trained in repairing vacuum-tube operated computers, or it may be impossible to find replacement parts for 16 Kb DRAM chips. Or the system may simply not incorporate the latest technology that in and of itself is seen desirable by users (for example, a rotary dial telephone system).
Economic Life (cost of operation). A system may still be functional, but become too expensive to be worth continuing to use. One example is because of a high cost of repair using obsolete components (this is a typical problem in long-lived embedded systems). Another reason may be that newer versions can be purchased and have lower operating costs so that the "payback" period of making that purchase is brief. This has, for example, happened recently with fuel-efficient furnaces and air conditioners.

Although it may not be possible to completely predict the lifetime of a system in advance, it is estimated taking these three factors into account. Then, the direct costs of ownership are considered, including:

Initial purchase cost. Clearly purchase cost is part of total cost. The usual issue is optimizing whether one should pay a higher up-front purchase cost in hopes of reaping lower operating costs. In some cases there is a limit on the amount of money that can be spent, such as a credit line limit, which may cap the allowable purchase cost.
Energy costs. Operating equipment usually requires energy, and can be a significant portion of total costs. In many cases embedded computers are used to increase energy consumption efficiency, and thus reducing energy cost is a primary goal. As an example, high-end home furnaces perform energy management to maximize heat delivered to the house and minimize heat that escapes into a basement from hot water "stranded" in the basement heating elements when the thermostat reaches its set point.
Maintenance/Repair/Custodial costs. A low initial purchase cost may be indicative of a system which will need frequent maintenance, repairs, or upkeep. Presumably a higher purchase cost indicates a system that contains more durable components. For an embedded system, it is more likely that components other than computers will break. However, installing sensors, data logging, and diagnostic capabilities for the system can substantially reduce these costs. As an example, elevators may come with minimal diagnostic sensors, but a contract maintenance company may well add sensors in an up-front investment to reduce the cost of later service calls.
Alteration/replacement costs. In a long-lived system that will be upgraded, it is important to take into account eventually removing or upgrading the equipment. As an example, a component or housing may be glued into place to save on installation costs, but be very difficult to remove, whereas a bolted-in system is more expensive to install, but cheaper to replace. A more specifically electronic example is the use of flash memory to permit field software upgrades without replacing read-only-memory chips.

Additionally, there are many indirect costs that must be taken into account in a complete financial model. These indirect costs of ownership include:

Interest (debt service). In some cases the most important indirect cost is the cost of borrowing money to pay the initial purchase cost in order to reduce later operating costs (or, alternately, the opportunity cost of not investing the purchase cost in some other way). Thus, any life cycle savings must be higher than extra initial cost savings to take into account the fact that extra money may need to be borrowed early in the system life cycle, but the savings are reaped later in the life cycle.
Administrative costs. These can vary considerably, but might include such factors as periodic safety inspections, the cost of arranging for and administering service agreements, the cost of tracking capital equipment via property tags, and the like.
Staffing of equipment to operate it. In some embedded systems computers are used to automate tasks previously performed by people. Perhaps the best everyday example of this is in elevators, which used to be operated by a trained person, but now are completely automatic. Clearly, saving operator wages is worth a significant increase in initial purchase cost in industrialized countries (but not necessarily in countries with low standards of living).
Opportunity cost of down time. A system that is frequently unavailable for service may not be as cost-effective as a more dependable system because of reduced productivity, the cost of stockpiling against potential service outages, or the cost of paying operators while their equipment is broken. Minimizing down time is extremely important most embedded system industries, including manufacturing and transportation. Embedded computers often contribute significantly to dependability by permitting operating in degraded modes while awaiting repair, by alerting maintenance personnel to impending failures, or by helping diagnose problems for quick repair.
Other non-quantitative factors. Some factors are important, but hard to set a fixed price on. These include potential productivity increases from comfortable employees and comfort taken from dealing with a company with a track record for standing behind their products.

The above life cycle costing approach works well for many large-scale embedded systems, where consumers are very sophisticated and equipped to invest for the long term. However, with consumer products it may be much more difficult to create a design that is optimized both to achieve profitable sales as well as maximum utility for the purchaser.

Antonides [Antonides90] discusses a combination of economics and psychology for durable goods purchases. While they come to the conclusion that frequency of use is generally related to reliability, and the cost of the good is generally related to its lifetime (when comparing different similar goods of different prices). While this is probably no surprise, the finding that makes life difficult for embedded system designers is that the decision of when to scrap a product is made on a potentially distorted view of life cycle economics. It should be noted that this is a European study and thus not necessarily representative of North American consumers. However, the specific points mentioned seem to ring true in general (the following is an interpretation and amplification, not a quotation):

The opinion of consumers with respect to how their equipment works does not agree with objective technical measurements. In particular, there is a bias to replacing rather than repairing, even when repair would be more economical in a life cycle cost sense.
Consumers, and especially low-income consumers, tend to underestimate the benefit of paying a higher purchase cost and reaping lower operating costs. This is presumably because they have little cash, and they have little credit or only have access to expensive credit such as credit cards rather than home equity loans. (Thus, many such systems are extremely cost-sensitive, squeezing the budget for embedded system components.)
The age of the consumer may affect the degree of patience and tendency to perform long-term planning (thus, younger or very old people may not value features such as diagnostics or upgradability; additionally they may not be inclined to wait and save for buying a high-end model).

When taken as a whole, the point of life cycle costing is to take into account all the direct and indirect costs of a product, and optimize for the lowest total cost given the constraints of customer preferences/behaviors.

Logistics

Logistics is the process of managing acquisition, movement, and storage of parts, materials, and products [Christopher92]. It has been a specialized discipline for a very long time in the military, which is of course vitally connected with keeping troops properly supplied. In the embedded systems life cycle, logistics affects the life cycle phases from production through retirement.

Logistics can be viewed along two axes: by type of item, and by activity. In either case the objective is to assure that materials are available where and when needed at minimum cost. This often involves tradeoffs of risks that demand or supplies will fluctuate vs. the cost of maintaining inventory (these costs include not only the cost to make space and staffing available, but also the debt burden of the money tied up in inventory as well). The particular types of items affecting logistics are:

Manufacturing components. Component flow must be assured to keep manufacturing lines working, but maintaining excessive inventory costs money. Many manufacturing plants in general have switched to "just in time delivery", in which they dramatically reduce inventory while depending on assurances from suppliers that component flow will be uninterrupted. Additionally, many products have undergone a process known as "design rationalization," in which inexpensive components may be replace with more expensive components or materials if doing so results in a lower global cost. An embedded system example might be using two different components at a quantity of 50,000 per year, where one component is $.01 cheaper than the other because of reduced performance, but otherwise they are identical. Using 100,000 per year of the more expensive component might yield a quantity price break of $.02 per component, lowering total cost.
In addition to these general techniques, embedded systems can exploit design features such as programmable memory instead of mask-programmed memory so that they use a lesser number of different components. These individual components may be expensive, but those costs may be offset by reduced inventory cost as well as flexibility to respond to sudden changes in product mix demands or changes in designs (for example, even high-volume auto component manufacturers are considering using flash memory instead of masked ROM because they receive frequent change requests from customers, resulting in costs for new masks and a need to potentially discard obsolete components that have not yet been used). An additional technique that can be used is incorporating field-programmable logic into designs to displace many different standard components with fewer types of generic components.
Finished products. Finish products must be stored and transported to customers. Changing designs to make completed systems simpler to store and transport might reduce logistics costs. For example, having a single generic system model that can be warehoused and customized at shipment time or by a product dealer may permit a single system to be manufactured and sold as several different products (this might be local language customization, or in some cases has been known to be two products with different feature sets and different prices that are in reality a single design).
Spare parts. There is a potentially very large opportunity in embedded systems to reduce costs for spare parts, but this is a difficult problem. Because technology becomes obsolete far more quickly than long-lived embedded systems are retired, it is common for support companies to be forced to make "end of life buys" for semiconductor components. They must therefore have money tied up for years or decades in components that may or may not ever be needed for eventual repairs. As an example, the auto industry is required by law to provide spare parts for a decade after a car is manufactured. The high cost of doing this in part accounts for the fact that spare parts are very expensive (so expensive, in fact, that building a car from spare parts is about 7 times more expensive in materials than the purchase price of a new car). Embedded system designers can help reduce this cost by attempting to design generic systems in which new-technology components can be plugged in to replace old ones; however this is still an open research area.

The types of activities that most affect logistics are below. They are to a large degree a different view on the same issues as the types of items, with the exception of disposal concerns:

Delivery. Embedded systems and components must be transported to the factory, to the customer, and to disposal.
Inventory. Components, systems, and spares must be kept in stock.
Disposal. Obsolete or broken systems must be disposed of. Systems may be refurbished, in which case they must be collected and shipped to the incoming inventory of a rework factory. They may be disassembled into constituent materials and put into the recycling waste flow, or they may simply be discarded. The discussion on Green Design below discusses embedded system strategies for this activity.

The main concern about logistics for embedded system designers is to ensure that design decisions are made on a more global scope than simply minimizing per-unit component costs. In particular, it is important to take into account the cost of sinking a large amount of money into spares that must be stored over many years for long-lived systems.

Life Cycle Assessment ("Green Design")

A third general area of life cycle considerations is specifically that of designing products for ecologically friendly retirement and disposal. This is a relatively new, but fairly active area of academic research, with roots back to the ecology movement of the 1960s. The basic idea is that products should be designed with disposal, operational consumption of resources, and operating pollution specifically in mind, just as manufacturing efficiency is considered in many designs. This area has now matured to the point that there are international standards for approaching the problem, including the ISO 14000 series, and in particular ISO 14040 "Life Cycle Assessment".

As an idea of the magnitude of the impact a system designer can have on recycling effectiveness, [Goldberg98] states that there is an IBM estimate that discarded computers will occupy 2 million tons of US landfill space by the year 2000. Although embedded systems are more prevalent, it is in some cases more difficult to understand their impact on landfill space because they are so often hidden inside other equipment. However, even so, there are many techniques that embedded system designers can employ to reduce the environmental impact and societal costs of their designs, including:

Design for repair. Some electronic assemblies are not designed for repair or replacement. While we live in a "throwaway" society, the question of whether something will actually be repaired is one that should be considered seriously rather than taken for granted.
Design for upgradability. In some cases embedded systems can be designed in a way that permits upgrading with new capabilities via exchange of electronic components. While this may add to initial cost, the company might in fact be able to make higher overall products by selling proprietary upgrade electronics rather than potentially losing the sale of a next-generation system to a competitor. Obvious upgrades for embedded systems include faster processors and larger memories to support added functionality. Of course, in some cases a simple capability for updating programming is sufficient to deliver upgrades.
Design to minimize power consumption. For battery-powered equipment this reduces the number of batteries that enters the landfill. (Remember that even rechargeable batteries have finite life, and may contain substances such as Cadmium that are quite toxic when disposed of.) Additionally, minimizing power reduces the consumption of fossil fuels, although in some larger embedded systems the power consumption of the rest of the equipment dwarfs electronic power consumption.
Design for recycling/clean disposal. This can include designing a system so that different material types are readily separated for recycling. It can also encompass using less toxic materials, such as non-lead bearing solder in circuit board manufacturing.
Design for clean manufacturing. Manufacturing uses resources and produces waste materials. A recent example of manufacturing becoming cleaner for embedded systems is the switch from Freon as a circuit board cleaner to aqueous cleaners (even rubbing alcohol and a scrub brush will work, but static discharge must be avoided).

In general, the Green Design approach to the life cycle is to consider the end of the product life when doing the design. A tricky way in which this ties in to the life cycle costing model is that society as a whole (and our descendents) pay a hidden price for pollution and resource exhaustion that are difficult to take into account in a single-product life cycle model. The only proven wide-scale techniques for accounting for these costs is government taxation/rationing (such as with Freon to discourage manufacture and consumption) and recent European initiatives to make manufacturers accountable for disposal of waste streams their products generate.

Available tools, techniques, and metrics

The state of the art in tools, techniques, and metrics varies widely depending on the particular area in life cycle concerns. Life cycle costing is generally an accounting approach that is not encountered by every-day design engineers, and measures results in dollars. Logisticians have a variety of mathematical optimization tools available, and in fact are intimately tied to the field of Operations Research (e.g., the Simplex method), and measures results in terms of achieving maximum efficiency for a given flow rate objective. Green design currently employs tools of a sort of "spreadsheet" nature that link raw materials and components with the waste streams they generate at end-of-life, and typically measures results in terms of tons of landfill space consumed.

Relationship to other topics

Life cycle concerns are not a particular technical expertise area the way that many of the other topics are. Instead, it is an overlay concept that suggests that each technical area consider the impact of decisions on the rest of the areas.

Conclusions

The following are the key ideas for this topic:

The term "life cycle" has several meanings in various communities, but they include a cost model for total cost of ownership (and/or total profit); dealing with the logistics of supplying, transporting, and maintaining equipment; and end-of life disposal/retirement issues.
A sufficiently rich economic model can help designers and users make informed decisions to minimize total cost of ownership. However, the real economic constraints and potentially suboptimal behavior of customers may lead to making choices for purchases and operation that are less than what would be optimal without these constraints.
A true life cycle perspective includes designing not only for manufacturing cost, but also includes the costs (monetary and otherwise) for all the phases of the lifecycle, including manufacturing, deployment, maintenance, operation, and eventual retirement/disposal.

Annotated References

[Antonides90] Antonides, Gerrit, The lifetime of a durable good : an economic psychological approach. Boston: Kluwer Academic, 1990.
This book combines economics and psychology to study the effect of people's attitudes and equipment characteristics on decisions to dispose of durable goods (in other words, the equipment lifetime based on people's behavior rather than on when the goods are "really" worn out). The study was done in the Netherlands.

[Christopher92] Christopher, Martin, Logistics: the strategic issues, London: Chapman & Hall, 1992.
A book with twenty papers spanning the range of how logistics affects manufacturing industries. Of particular interest for embedded systems are Chapter 16 about General Motors and Chapter 18 about BMW.

[Goldberg98] Goldberg, L., "The advent of 'green' computer design." IEEE Computer, Sept. 1998, vol.31, no.9, p. 16-19.
This is a general introduction to design-for-disposal, using an example of personal computers to illustrate the issues.

[Kirk95] Kirk, Stephen J.; Dell'Isola, Alphonse J., Life Cycle Costing for Design Professionals. New York: McGraw-Hill, 1995.
This is a thorough text on the economics/accounting/finances of equipment purchasing with respect to life cycle cost optimization.