Identifying and Understanding General Purpose Technologies:

Which technologies have rapid improvements and why?

Jeffrey L. Funk

Associate Professor

National University of Singapore

Division of Engineering and Technology Management (EA-5-34)

9 Engineering Drive 1, Singapore 117576

etmfjl@nus.edu.sg

Identifying and Understanding General Purpose Technologies:

Which technologies have rapid improvements and why?

Abstract

This paper helps economists, policy makers and firms identify and understand general purpose

technologies and the rapid rates of improvement that are a necessary characteristic of them. First

it uses a published data base on annual rates of improvement for 120 technologies to show that

67% and 89% of the technologies experience rates of improvement of less than 9% and 15%

respectively and that the data contains multiple distributions. Second, it presents a separate data

base of technologies that have annual rates greater than 25% and thus can be considered candidates

for general purpose technologies. Third, it analyzes a number of potential explanations for rapid

rates of improvement and concludes that greater opportunities for improvements is the most likely

reason. Fourth, it describes two types of product and process design changes that can explain

annual rates of improvement that are greater than 25% and that are more applicable to some

technologies than to others.

1. Introduction

Many economic scholars argue that general purpose technologies (GPTs) have a large impact

on the productivity growth of advanced nations that goes beyond the impact of typical innovations,

which by themselves have a large impact on productivity growth (Solow, 1957). Technologies

such as steam engines, railways, internal combustion engines, motor vehicles, integrated circuits,

computers, lasers, and the Internet have been defined as GPTs due to their large impact on

productivity growth and thus standard of living (David, 1989; Bresnahan and Trajtenberg, 1995;

Helpman, 2003; Lipsey et al, 2005). For example, there is a growing number of papers that have

analyzed the relationship between computers and economic growth (Oliner and Sichel, 2002;

Olner, Sichel and Stiroh, 2007; Jorgensen et al, 2008) where it is recognized that improvements in

integrated circuits are the sources of the improvement in computers by both economists (Bresnahan

and Trajtenberg, 1995) and computer scientists (Smith, 1989). Although several definitions of

GPTs have been provided, rapid rates of improvement for a technological trajectory (Dosi, 1982)

are a common theme among these definitions along with the applicability of a technology to a

wide variety of applications (Bresnahan and Trajtenberg, 1995; Lipsey et al, 1998).

However, few economic or management scholars have addressed, even indirectly, how rapid

rates of improvement might occur (exceptions include: Gold, 1974; Lipsey et al, 2005; Winter,

2008; Funk, 2013a) and similarly few have attempted to present data on rates of improvement

(Funk, 2013b; Nagy et al, 2013). Such data and analyses would be an essential part of identifying

potential GPTs and such an identification would probably enable better R&D and other economic

policies. For example, reducing carbon emissions through less usage of fossil fuels requires new

forms of GPTs. However, these GPTs must have rapid rates of improvement in performance and

cost in addition to low carbon emissions in order that they will become economical for a wide

variety of applications. This requirement should probably exclude wind turbines and electric

vehicles, which are emphasized by the Intergovernmental Panel on Climate Change (IPCC, 2013),

from a list of carbon-reducing GPTs since they have slow rates of improvement (IPCC, 2013;

Tarascon, 2009).

This is probably the first economic paper to analyze a large data base on rates of improvement

and how rapid improvements occur. Section 2 analyzes existing annual rates of improvement for

120 technologies from Nagy et al (2013) and shows that the data contains multiple distributions.

Annual rates between 15% and 25% contain no technologies and 67% and 89% of the technologies

have annual rates of improvement of less than 9% and 15% respectively. Furthermore, the small

literature on the economics of improvements suggests that the technologies with annual rates of

improvement of less than 15% can probably be further divided into two normal distributions, one

for fixed formula chemicals and one for other technologies. The rates of improvement for the fixed

formula chemicals are almost twice as high as are the other technologies probably because they

benefit more from economies of scale than do the other technologies.

Section 3 presents data on technologies with recent annual rates of improvement that are greater

than 25% per year. Although 25% per year is a much higher rate of improvement than some GPTs

such as steam engines and railways experienced in the 18th and 19th centuries, most scholars agree

that rates of change in the late 20th century have been much higher than in previous centuries so

we set a rather high rate of 25% per year as a cutoff for defining rapid rates of improvement. These

technologies can be considered potential GPTs and their becoming GPTs probably depends on

whether these rapid rates continue and whether the breadth of their applications becomes

sufficiently broad.

The fourth section examines several explanations for why some technologies experience annual

rates of improvement that are greater than 25% per year. These include greater: 1) production; 2)

demand; 3) R&D; and 4) opportunities for improvements. Consistent with previous research that

focused on inter-industry differences in R&D (Rosenberg, 1974; Klevorick et al, 1995), we

conclude that greater opportunities for improvements is the largest reason for the differences in

rates of improvement.

The fifth section examines reasons for the greater number of opportunities for improvements in

some technologies than others. While previous research has focused on the impact of science on

specific industries or technological advances originating outside these industries (Klevorick et al,

1995), this paper focuses on product and process design changes that can explain rapid rates of

improvement and that are more applicable to some technologies than to others. We conclude that

two types of product and process design changes, scaling and materials creation, can enable rapid

rates of improvement and are applicable to some technologies more than to others.

2. The Distribution of Rates of Improvement

This section presents and analyzes data on rates of improvement. Data on rates of improvement

are calculated from performance vs. time curves that are presented in Nagy et al (2013) and

publically available on a website (http://pcdb.santafe.edu/). We use all of the technologies that are

presented on this website. We calculate rates of improvement from the changes in performance

and/or cost between the first and last data points and from the number of years separating the first

and last data points.

Figure 1 plots the number of technologies as a function of annual rates of improvement in a bar

graph. The percentages along the x-axis are defined as plus or minus one percentage point except

for those with annual rates of improvement exceeding 42%. Thus, the 9 technologies shown for a

10% annual rate of improvement have rates that are between 9.0 and 10.9%. Excluding the ones

with annual rates greater than 42%, the mean for the 120 technologies is about 5% and the standard

deviation is about ….

Most of the technologies have slow rates of improvement and two have negative rates. Sixty

five and 89% of them have annual rates of improvement that are less than 9% and 15%

respectively, and the rest have rates of greater than 25% per year. The gap between annual rates of

15% and 25% suggest that multiple distributions exist within Figure 1 and these multiple

distributions may illuminate how these improvements occur and thus how to better identify

potential GPTs.

What types of multiple distributions might exist in Figure 1? In addition to the technologies with

annual rates less than 15% and greater than 25%, there also may be multiple distributions within

the main group of technologies that have rates lower than 15%. One hypothesis is that chemicals

may constitute a separate group of technologies since they have unique characteristics that

probably impact on their rates of improvement. Their fixed formulas limit the possible changes in

product design (Stobaugh, 1988), they benefit from increases in the scale of production equipment

more than do other technologies (Pratten, 1971; Chandler, 1994; Funk, 2013a; Funk, 2013b), and

the relationship between the capital cost of their production facilities and their volumes is highly

regular. The capital cost for these chemical factories is a function of plant size to the nth power

where n is typically between 0.6 and 0.7 (Haldi and Whitcomb, 1967; Axelrod et al, 1968;

Rosenberg, 1994; Mannan, 2005).

The hypothesized reason for this highly regular relationship between capital cost and plant size

is due to the impact of geometry on costs; the cost of pipes varies as a function of radius while the

output from pipes varies as a function of radius squared. Similarly, the costs of reaction vessels

vary as a function of surface area (radius squared) while the output of a reaction vessel varies as a

function of radius cubed (Rosenberg, 1994; Winter, 2008; Funk, 2013a: Funk, 2013b). This causes

the capital costs of their production facilities to fall in direct proportion to increases in production

volumes.

We investigate the possibility that chemicals constitute a separate distribution by plotting the

number of chemical and non-chemical technologies vs. rate of improvement in Figures 2 and 3.

Both of these figures show normal distributions with different means and standard deviations. The

chemical technologies have an average annual improvement rate of 6.3% while the non-chemical

technologies have an average annual improvement rate of 4.1%. For the chemical technologies,

the differences in annual rates among technologies in Figure 2 are probably from differences in

the role of material costs and from differences in timing. Since the increases in production volume

impact on capital costs and not materials cost (unless the increased volumes impact on upstream

activities), chemicals with higher materials costs will experience slower rates of cost reduction.

Differences in timing will result from the fact that the benefits from increases in scale are probably

larger during early years when the scale-up is first started.

For the non-chemical technologies, it is interesting that these technologies have lower annual

rates of improvement than do the chemical technologies in spite of having more opportunities for

product design changes than do the chemical technologies. Non-chemical technologies have

greater opportunities for improvements through changes in product design since they don’t have

fixed chemical formulas as chemicals do (Stobaugh, 1988) and thus one would expect more rapid

rates of improvement for them than for chemicals. Since they have slower rates of improvement

than do chemicals, there must be much fewer benefits from increasing the scale of production

equipment for non-chemicals than from chemicals. This conclusion is consistent with previous

research (Pratten, 1971; Chandler, 1994; Funk, 2013a; Funk, 2013b).

3. Technologies with Rapid Rates of Improvement

This section presents data on technologies with annual rates of improvement that are greater

than 25% per year. Our analysis focused on technologies with current rapid rates of improvement

since we are interested in identifying potential GPTs and since the definition of rapid has probably

changed over time. Many argue that technology change is much more rapid in the late 20th than in

previous centuries and the rapid rates of improvement that are presented in this paper are consistent

with an acceleration in technology change. For example, rates of improvement for many electronic

technologies are much more rapid than were those for GPTs such as internal combustion engines

and other power delivery systems in the early 20th century, as reported in Nagy et al (2013).

We collected the data on annual rates of improvement from scientific and engineering journals

such as Nature, Science, Phys Status Solidi and IEEE, annual reports by reputable scientific

organizations such as the International Solid State Circuits Conference, archival social science

publications (Martino, 1971; Koh and Magee, 2006; Koh and Magee, 2008), and general

technology and technology-specific web sites. All of the data are from a single source except DNA

sequencing, which includes data from two sources (Singularity, 2014; NHGRI, 2014). Since a

variety of performance measures are often relevant for a specific technology, data was collected

on multiple dimensions when possible, some of which are represented in performance per unit cost

while others are in performance per mass or per volume. Some of the performance data are for

best laboratory results, some of them are for commercialized products, and some of them combine

best laboratory results for early prototypes with best products for after commercialization occurs.

In all cases, we used the raw data from the original sources, thus relying on the reputations of these

sources.

Technologies with annual rates of improvement greater than 25% are shown in Table 1. To aid

the reader, the technologies are placed into 8 categories. The first six categories are the

transforming, storing, and transporting of energy and information, and the last two are for the

transformation and transportation of living organisms. As expected many of the technologies in

Table 1 can be defined as information technologies.

All of the technologies shown in Table 1 can be considered either existing GPTs or potential

GPTs. The former probably includes various types of ICs (microprocessors, flash memory,

DRAMs), computers, and both wireline and wireless transmission systems. The other technologies

may become GPTs if their breadth of applications dramatically increases, which some may.

Figure 4 shows the number of technologies with rates of improvement that are between the

specific rates of improvement on the x-axis. For example, there are 7 technologies with annual

rates of improvement between 25% and 35% and 6 technologies with annual rates of improvement

between 35% and 45%. The number of technologies declines as the rates of improvement increase.

The next section focuses on the reasons for rapid rates of improvement.

4. Why do some technologies have very rapid rates?

This section addresses four possible reasons for the more rapid rates of improvement for the

technologies in Table 1. They are greater production volumes, demand, R&D, or opportunities for

improvements.

4.1 Greater production volumes

Analyses of cost by the economics literature primarily focus on the factory floor and links cost

reductions with cumulative production; an exception is (Thompson, 2012). In what has been

termed learning by doing (Arrow, 1962), costs fall as firms learn to produce a single design in a

single factory more efficiently and thus with lower costs. Workers become better at tasks and firms

introduce better work flows, (Wright, 1936, Argote and Epple, 1990; Adler and Clark, 1991;

Thornton and Thompson, 2001), better process control (Argote, 1999; Lapre et al, 2000), and

automated manufacturing equipment (Utterback, 1994), and they promote organizational learning

(Benkhard, 2000). Thus, some technologies might experience faster rates of improvement merely

because they have a faster growth in production volumes.

We give two reasons why greater production volumes are not the main reason for the rapid rates

of improvement in the technologies shown in Table 1, in addition to the reasons cited by Thompson

(2012). First, we note that a previous section contrasted the rates of improvement for chemicals

and non-chemicals, both within the technologies experiencing annual rates less than 15%. That

section showed that chemicals have more rapid rates of annual improvements than do non-

chemical technologies in spite of the fewer opportunities for product design changes in chemicals

than in non-chemicals. That analysis and previous research (Pratten, 1971; Chandler, 1994;

Rosenberg, 1994; Winter, 2008; Funk, 2013a: Funk, 2013b) suggests that this is probably because

chemicals benefit more from increases in the scale of production equipment, which is one form of

learning by doing and it requires increases in production volumes. Nevertheless, since chemicals

do not experience rates of improvement that approach those of the technologies shown in Table 1,

it is unlikely that rapid rates of improvement in Table 1 are due to greater production volumes.

A second and perhaps more important reason why greater production volumes don’t explain the

rapid rates of improvement in Table 1 is that many of the dimensions are performance as opposed

to cost-related measures. Improvements in performance probably require product design changes

(Utterback, 1994) and thus increases in production don’t directly lead to reductions in cost. On the

other hand, they might indirectly lead to improvements in performance since increases in

production is associated with increases in demand and increases in demand often lead to increases

in R&D spending.

4.2 Greater Demand

A second possible reason for faster rates of improvement is greater demand. We note that this

is actually a partial variant of learning curves since demand is related to cumulative production.

The difference between the impact of demand and production on costs is that increasing demand

can induce greater R&D (Schmookler, 1966) and thus lead to lower costs and higher performance

through the greater R&D. This has caused some scholars to consider cumulative production as a

general proxy for effort and thus the driver of new product and process designs and improvements

in performance and cost (Lieberman, 1984; Dutton and Thomas, 1984; Balasubramania and

Lieberman, 2010).

We give two reasons why we believe this is an inadequate explanation for the differences in

rates of improvement. First, we note that many of the technologies listed in Table 1 experienced

rapid improvements without commercial production and many of them still do not have large

amounts of commercial production and thus markets. This includes organic LEDs and transistors,

quantum dot solar cells and displays, Perovskite solar cells, superconducting Josephson junctions,

Quantum computers, carbon-nano tubes for transistors, and new forms of non-volatile memory

such as Resistive RAM Ferroelectric RAM, Magneto RAM, and Phase Change RAM (random

access memories). These technologies have experienced rapid improvements without commercial

production and thus without demand (Funk and Magee, 2014).

On the other hand, one might argue that there is strong demand for research on these

technologies although the demand is not yet seen in viable markets. This was also part of

Schmookler’s (1966) explanation for innovation in that he argued that demand existed for the

important innovations that preceded his research. This brings us to the second reason why we

believe that demand is an inadequate explanation for rapid improvements. This explanation is best

summarized by looking at Rosenberg’s (1974) response to Schmookler’s (1966) argument.

Rosenberg (1974) response was that “some demand induces the inventions that satisfy it. But

which and when?” He then described many areas in which demand has existed for many years if

not centuries (e.g., health care), but was not satisfied until recently. He concludes this paper with:

“the economic question is: Given the state of the sciences, at what cost can a technological end be

attained?” We offer a possible answer to Rosenberg’s question in sub-section 4.3 and section 5.

4.3 Greater R&D Intensity

A third possible reason for the more rapid rates of improvement for the technologies in Table 1

is greater R&D intensity. Perhaps more R&D has been devoted to these technologies than other

technologies and this has caused the technologies in Table 1 to have more rapid rates of

improvement than do other technologies. Furthermore, this greater R&D might come from greater

production, as mentioned in the previous sub-section.

Research on inter-industry differences in R&D suggest that this is not an adequate explanation

for the rapid rates of improvement in Table 1. Research on inter-industry differences in R&D focus

on two issues - differences in appropriability and differences in opportunities – of which the second

reason is generally perceived to be the more important reason (Klevorick et al, 1995). Although

differences in appropriability do impact on firm R&D, the sign is not clear. While strong

appropriability increases incentives to engage in R&D, weak appropriability lowers the cost of

research for others and thus differences in appropriability does not probably explain the rapid rates

of improvement in Table 1.

Differences in opportunities probably explains these differences in inter-industry R&D

(Klevorick et al, 1995), a conclusion that was reached twenty years earlier by Rosenberg (1974).

Klevorick et al (1995) based their conclusions on a survey of R&D managers. The survey asked

managers to assess the impact of various fields of science on technological progress and about

technological advances originating outside their industry. From this survey Klevorick et al (1995)

concluded that differences in opportunities best explain the differences in R&D intensity among

firms. Their conclusions is consistent with this paper’s analysis. The one difference between our

conclusion and Kleverick et al’s conclusions is that there other ways of measuring differences in

opportunities for R&D in addition to the method used by Klevorick et al (1995) and the next sub-

section and section describe a different approach to explaining differences in opportunities than

with surveys of R&D managers.

4.4 Greater opportunities for improvements

Building from the previous sub-section (Rosenberg, 1974; Klevorick et al, 1995), a third

possible reason for the rapid rates of improvement in Table 1 is greater opportunities for

improvements through greater product and process R&D. R&D leads to more improvements per

R&D dollar for some technologies than for other technologies and these improvements lead to

increased consumer welfare, profits, and R&D spending. Thus, the technologies that are easier to

improve end up having faster rates of improvement where greater R&D on these technologies

magnifies the differences in rates of improvement among different technologies. This can lead to

the dramatic differences between technologies that are shown in Figure 1.

This argument is a variant of Klevorick et al’s (1995) argument. They explain different levels

of opportunities in terms of the applicability of science to technological progress and the

technological advances originating outside their industry and they measure these factors through

surveys. This paper focuses on rates of improvement where the applicability of science or

technologies advances originating outside the industry might impact on the rates of improvement.

Thus, it could be that Klevorick et al’s (1995) factors might be the ultimate explanation for

differences in rates of improvement but this paper provides a more detailed analysis of the product

and process design changes that may be supported by Klevorick et al’s (1995) factors.

5. What kinds of design changes provide opportunities for rapid improvements?

A few scholars have attempted to understand the types of product and process design changes

that lead to improvements in cost and performance. Increases in the scale of production equipment

has already been mentioned and it is probably the most widely accepted mechanisms for costs

falling over time. Increases in the scale (e.g., diameter) of pipes and reaction vessels lead to lower

costs of chemicals in a highly regular manner because costs rise slower than do volumes as the

diameter is increased.

Some scholars call this geometric scaling (Winter, 2008). Geometric scaling refers to the

relationship between the geometry of a technology, the scale of it, and the physical laws that govern

it (Funk, 2013). Or as others describe it: the “scale effects are permanently embedded in the

geometry and the physical nature of the world in which we live” (Lipsey at al, 2005). The physical

nature of the world we live includes steam engines, internal combustion engines, electricity

generating plants, oil tankers, and freighters. All of these technologies benefit from increases in

scale in much the same way that chemical plants do (Lipsey et al, 2005) and some of them have

been defined as general-purpose technologies (David, 1989; Bresnahan, 1995; Helpman, 2003;

Lipsey et al, 2005).

The concept of geometric scaling can also help us understand when technologies benefit from

reductions in scale. As others have noted (Winter, 2008; Funk, 2013a; Funk, 2013b), reducing the

scale of features can lead to dramatic improvements in performance and price that go far beyond

the benefits from increases in scale. The reason is because making something smaller almost

always leads to long-term reductions in material, equipment, and transportation costs whereas the

technologies benefiting from increases in scale only get cheaper because costs don’t go up as fast

as output does. Thus, geometrical scaling and reductions in scale can explain many of the

technologies shown in Table 1.

The key issue in determining whether a technology benefits from reductions in scale is whether

some aspect of performance rises with reductions in scale. The performance of very few

technologies benefit from reductions in scale. But many of them experience very rapid rates of

improvement and most of them are related to information transformation, storage, and

transmission. Looking at Table 1, reducing the size of transistors in a microprocessor (Ferain et al,

2011), of pixels on a camera chip (Suzuki, 2010), of electrodes in a MEMS-based artificial eye

(Chader et al, 2009) or inkjet printer (Stasiak, et al, 2009), of devices in photonic chips (Evans et

al, 2011), and of memory cells in flash memory (Sanvido et al, 2008), resistive RAM, ferroelectric

RAM, magneto RAM, phase change RAM (ISSCC, 2013), magnetic tape or disks, and of the

processing function in DNA sequencers and synthesizers (Margulies et al, 2005; Ryan et al, 2007)

lead to rapid improvements in cost and/or performance. In all of these technologies, the reductions

in scale involve changes in both product and process designs. Furthermore, the resulting rapid rates

of improvements in these technologies lead to rapid rates of improvement in higher-level systems

that are shown in Table 1; these include digital and quantum computers, wireless cellular, LAN

(local area networks), and even wireline telecommunications to some extent (Funk, 2013a; Funk,

2013b).

A second type of product-process design change that can lead to rapid rates of improvement

involves creating materials (and their associated processes) that better exploit a physical

phenomenon (Funk, 2013a; Funk, 2013b). This type of design change can explain the rapid rates

of improvement that are not driven by reductions in scale and it has contributed to some of the

rapid rates of improvement that are primarily driven by reductions in scale. In the creation of these

new materials, there is a tight linkage between creating materials and the processes for making

them.

Beginning with LEDs (light emitting diodes) and organic LEDs (OLEDs), scientists and

engineers reduced their costs by finding materials that better exploit the phenomenon of

electroluminescence, partly because these new materials also led to improvements in luminosity

per Watt. They found new combinations of semiconducting materials such as gallium, arsenide,

phosphorus, indium, and selenium for LEDs (Azevedo et al, 2009) and new combination of organic

materials such as nitrides and polymers for organic OLEDs (Sheats et al, 1996; Lee, 2005). Many

of the improvements in semiconductor LEDs also led to improvements in semiconductor lasers,

due to the similarities between them. Furthermore, better lasers also required better materials for

the heat sinks, mirrors, and bonders that are part of a laser package (Martinson, 2007; Orton, 2005).

Finding or creating new materials also enabled many of the improvements in the performance

and cost of quantum dot displays and solar cells, superconducting transmission applications, and

organic transistors. Types of materials that are mentioned as contributing to the improvements in

quantum dot displays and solar cells include conventional semiconductors such as silicon or

indium arsenide, more complex compositions (i.e., alloys), and selenide or sulfides of metals (Bae

et al, 2010; Kwak et al, 2012; Patel, 2011; Chandler, 2013). For superconductors, new materials,

including ones to package the superconductors, and their associated processes enabled

improvements in current-length per cost and the product of current and length (Shiohara et al,

2013; Paranthaman and Izumi, 2004). For organic transistors, it was new materials (Dong et al,

2010) along with better crystal structures and interfaces (Horiwitza, 2011).

A third type of improvement comes primarily from processes and this type of improvement is

really a subset of the second one and it is consistent with research on chemicals (Sinclair et al,

2000) that identified new processes arising from research as a major source of cost reductions. For

carbon nanotubes, the efforts were aimed at improving their purity and density through post-

synthesizing efforts, selective growth with predetermined properties, and the growth of CNTs on

new types of substrates (Franklin, 2013). For superconducting Josephson junctions, improvements

were achieved by creating new types of qubit structures and new processes for making these

structures (Devoret and Schoelkopf, 2013) where the new structures include different sizes and

orientations of tunnel junctions, superinductors, and resonators and new processes include

exposure to microwave radiation (Hardesty, 2011).

Why do some technologies experience more rapid rates of improvement through these three

types of design changes than do other technologies? The first type of design change is only

applicable to some technologies so by itself this type of design change can explain rapid rates of

improvement and thus help us identify potential general purpose technologies. However, it is less

clear why the second and third types of design changes would be more applicable to some

technologies than to others.

Interviews with several physicists suggest that technologies involving the deposition of thin

crystalline films of less than one micro-meter on other crystalline materials are easier to improve

than are other materials. All of the technologies in Table 1 that benefited from the second and third

types of design changes involve thin-film deposition of crystalline materials while other

technologies with slower rates of improvement do not involve thin film deposition of crystalline

materials. For example, crop yields, the magnetic strength of most magnets, the strength of most

materials, and the energy storage density of batteries and flywheels do not involve thin film

deposition of crystalline materials and their rates of improvement are less than 10% per year.

6. Discussion

Identifying and understanding general purpose technologies (GPTs) is an important task for

economists, policy makers and firms. The large impact that they have on economic growth suggest

that policies should favor their development over the other technologies. Firms are also interested

in such technologies since the resulting large markets probably suggest large potential profits.

Identifying them early in their development, however, remains problematic. One key

characteristic of them is a rapid rate of improvement and this paper may be the first economics

paper to present and analyze data on rates of improvement. It showed that 67% and 89% of the

technologies had annual rates of improvement that are less than 9% and 15% respectively and that

there is a gap between annual rates of 15% and 25%, thus suggesting multiple distributions exist

in the data. In addition to the two distributions suggested by the gap between annual rates of 15%

and 25%, this paper’s analysis also suggests that the technologies with annual rates of

improvement below 15% can be further divided into two distributions, one for chemical and one

for non-chemical technologies.

The paper then analyzed possible reasons for the rapid rates of improvement, defined as greater

than 25% per year. To do this, data was gathered and presented for technologies with these rapid

rates and argued that these technologies are potential candidates for general purpose technologies.

It then discussed possible reasons for the rapid rates of improvement, which included greater

production volumes, demand, R&D intensities, and opportunities for improvements. This

discussion concluded that greater opportunities for improvements are the probable reason for the

rapid rates of improvement for some technologies.

Finally the paper proposed three types of design changes that may explain the greater

opportunities for improvement and thus the rapid rates of improvement: 1) reductions in scale; 2)

creation of materials; 3) new processes. We concluded that by itself reductions in scale can explain

rapid rates of improvement since few technologies benefit from reductions in scale, mostly

information technologies, and all of them appear to have rapid rates of improvement. We also

concluded that the other two types of design changes might explain rapid rates of improvement in

the cases of technologies that involve the deposition of thin film crystalline materials. Further

research is needed in this area.

Further research is also needed on other characteristics of general purpose technologies, such as

the breadth of applications. Is there a way to a priori understand why some technologies have a

wider breadth of applications than do other technologies? Perhaps rapid rates of improvement also

impact on breadth if the rapid rates continue for many decades. Or is there some other characteristic

that is important. Further research is needed on this issue.

8. References

Adler P and Clark K 1991. Behind the Learning Curve: A Sketch of the Learning Process,

Management Science 37(3): 267-281.

Argote L 1999. Organizational Learning: Creating, Retaining and Transferring Knowledge,

NY: Springer.

Argote L and Epple D 1990. Learning Curves in Manufacturing, Science 247(4945): 920- 924.

Arrow K 1962. The economic implications of learning by doing, The review of economic

Studies 29(3): 155-173.

Azevedo I, Morgan G, Morgan F, 2009. The Transition to Solid State Lighting, Proceedings of

the IEEE 97: 481-510.

Bae W, Kwak J, Lim J, Lee D, Nam, M. Char, K, Lee C, Lee S. 2010. Multicolored Light-

Emitting Diodes Based on All-Quantum-Dot Multilayer Films Using Layer-by-Layer

Assembly Method. Nano Letters 10: 2368-2373.

Balasubramanian N and Lieberman M 2010. Industry Learning Environments and the

Heterogeneity of Firm Performance, Strategic Management Journal 31: 390-412.

Benkard L 2000. Learning and Forgetting: The Dynamics of Aircraft Production, American

Economic Review 90(4): 1034–1054.

Bresnahan T and Trajtenberg M, 1995. General purpose technologies 'Engines of growth?’,

Journal of Econometrics, Elsevier 65(1): 83-108.

Brown R 2011. Staying with Copper in a Fiber World, Rural Telecom, March-April: 22-26.

Chader G, Weiland J, Humayun M 2009. Artificial Vision: needs, functioning, and testing of a

retinal electronic prosthesis, Progress in Brain Research 175: 317-331

Chandler, A 1994, Scale and Scope: The Dynamics of Industrial Capitalism, Boston: Belknap.

Chandler D 2013. New solar-cell design based on dots and wires, MIT News, March 24, 2013.

David P 1990 The Dynamo and the Computer: An Historical Perspective on the Modern

Productivity Paradox, American Economic Review 80(2): 355-61

Devoret M and Schoeldopf R 2013. Superconducting Circuits for Quantum Information, Science

339, March 8: 1169-1174

Dong H, Wang C, Hu W 2010. High Performance Organic Semiconductors for Field-Effect

Transistor, Chemical Communications 46: 5211-5222

Dosi, G. 1982. Technological paradigms and technological trajectories, Research Policy 11 (3):

147-162.

Dutton J and Thomas A 1984; Treating Progress Functions as a Managerial Opportunity,

Academy of Management Review 9(2): 235-247.

Evans P et al 2011. 1.12 Tb/s superchannel coherent PM-QPSK InP transmitter photonic

integrated circuit (PIC), Optics Express 19(26): B154-B158

Ferain I, Colinge C, Colinge J 2011. Multigate transistors as the future of classical metal–oxide–

semiconductor field-effect transistors, Nature 479, 310–316.

Francis G 2011. Data Storage – Trends and Directions

http://lib.stanford.edu/files/pasig-jan2012/11B7%20Francis%20PASIG_2011_Francis_final.pdf;

Franklin A 2013. Electronics: The road to carbon nanotube transistors, Nature 498: 443-444

Funk J 2013a. Technology Change and the Rise of New Industries, Stanford University Press

Funk J 2013b. What Drives Exponential Improvements? California Management Review, Spring

Gold, B. 1974. Evaluating Scale Economies: The Case of Japanese Blast Furnaces, The

Journal of Industrial Economics 23(1): 1-18

Haitz R and Tsao J 2011. Solid State lighting: ‘The case’ 10 years after and future prospects,

Phys Status Solidi A 208(1): 17-29.

Hardesty L 2011. Long Live the Qubit, MIT News, 1 June 2011.

Hasegawa T and Takeya J 2009. Organic field-effect transistors using single crystals, Science

and Technology of Advanced Materials 10, 024314

Helpman E 2003. General Purpose Technologies and Economic Growth, MIT Press.

Horowitza 2011. The organic transistor: state-of-the-art and outlook, European. Physics Journal

Applied Physics 53 (33602)

IPCC, 2013. Renewable Energy Sources and Climate Change Mitigation: Special Report of the

Intergovernmental Panel on Climate Change. Cambridge University Press. 2013

ISSCC, 2013. International Solid State Circuits Conference, ISSCC Trends

Jorgenson D, Ho M, and Stiroh K 2008. A Retrospective Look at the U.S. Productivity Growth

Resurgence, Journal of Economic Perspectives 22(1): 3-24.

Klevorick A, Levin R, Nelson R, Winter S 1995, on the sources and significance of

interindustry differences in technological opportunities, research policy 24: 185-205

Koh, H. and Magee, C. 2006. A functional approach for studying technological progress:

Application to information technologies, Technological Forecasting and Social Change

73: 1061-1083.

Koh, H. and Magee, C. 2008. A functional approach for studying technological progress:

Extension to energy technology, Technological Forecasting and Social Change 75:

735-758.

Koomey J, Berard S, Sanchez M, Wong H, 2011. Implications of Historical Trends in the

Electrical Efficiency of Computing, IEEE Annals of the History of Computing 33(3): 46-54.

Kwak J 2010. PhD Thesis, cited in Changhee Lee, Seoul National University

http://www.andrew.cmu.edu/org/nanotechnology-forum/Forum_7/Presentation/CH_Lee.pdf

Kwak J, Bae W, Lee D, Park I, Lim J, Park, M, Cho H, Woo H, Yoon D, Char K, Lee S, Lee C

2012. Bright and Efficient Full-Color Colloidal Quantum Dot Light-Emitting Diodes Using an

Inverted Device Structure, Nano Letters 12: 2362−2366

Lapre M, Mukherjee A, Wassenhove L 2000. Behind the Learning Curve: Linking Learning

Activities to Waste Reduction, Management Science 46(5):597-611.

Lee, C 2005. OLED 1 – Introduction,

http://wenku.baidu.com/view/783fa93283c4bb4cf7ecd196.html

Lieberman M 1984. The learning curve and pricing the chemical processing industries. Rand

Journal of Economics 15: 213-228. R&D expenditures and capital intensity

Lipsey, R. Carlaw, K. and Bekar, C. 1998. What Requires Explanation? In General Purpose

Technologies and Economic Growth, Helpman (ed), MIT Press.

Lipsey, R. Carlaw, K. and Bekar, C. 2005. Economic Transformations, NY: Oxford Univ Press.

Mannan, S. 2005. Lee’s Loss Prevention in the Process Industries, Vol. 1, Burlington, MA:

Elsevier Butterworth-Heinemann.

Margulies M (and more than 20 co-authors) 2005. Genome sequencing in microfabricated

high-density picolitre reactors, Nature. 2005 September 15; 437(7057): 376–380.

Martino J 1971. Examples of Technological Trend Forecasting for Research and

Development Planning. Technological Forecasting and Social Change 2: 247-260

Martinson R 2007. Industrial markets beckon for high-power diode lasers, Optics, October: 26-

27. ww.nlight.net/nlight-files/file/articles/OLE%2010.2007_Industrial%20markets...pdf

Nagy B, Farmer D, Bui Q, Trancik J 2013. Statistical Basis for Predicting Technological

Progress. PLoS ONE 8(2): e52669. doi:10.1371/journal.pone.0052669NREL, 2013.

NHGRI, 2013. National Human Genome Research Institute.

www.genome.gov/sequencingcosts/, last accessed on September 10, 2013.

NREL, 2013. National Research Energy Lab.

http://en.wikipedia.org/wiki/File:PVeff(rev131204)a.jpg

Oliner S and Sichel D 2002. Information technology and productivity: where are we now and

where are we going? Economic Review 91: 1-32

Oliner S, Sichel D, Stiroh K 2007. Explaining a Productive Decade, Finance and Economics

Discussion Series 2007-63,

Orton J 2005. Semiconductors and the Information Revolution: Magic Crystals that made IT

Happen, Oxford University Press.

Paranthaman P and Izumi T 2004. High-Performance YBCO-Coated Superconductor Wires,

MRS Bulletin, August 533-536.

Patel, P 2011. Quantum Dots as Solar Cells, MIT Technology Review.

Pratten, C. 1971.Economies of Scale in Manufacturing Industry, Cambridge University Press.

Preil M 2012. Optimum Dose for EUV: Technical vs. Economic Drivers, April 26. Future Fab

41

Rosenberg N 1974. Science, Invention, and Economic Growth, Economic Journal 84: 90-108.

Ryan D, Rahimi M, Lund J, Mehta R, Parviz BA, 2007. Toward nanoscale genome sequencing.

Trends in Biotechnology 25(9):385-9.

Sanvido M, Chu F, Kulkarni A, Selinger R 2008. NAND Flash Memory and its Role in Storage

Architectures, Proceedings of the IEEE, November: 1864-1874.

Schmookler, J. 1966. Invention and Economic Growth, Cambridge, Harvard University Press.

Selvamanickam V 2011. Second-generation HTS Wire for Wind Energy Applications,

Symposium on Superconducting Devices for Wind Energy, 25 February, Barcelona, Spain

Sheats, J, H Antoniadis, M Hueschen, W Leonard, J Miller, R Moon, D Roitman, A Stockinget,

Organic Electroluminescent Devices, Science 273 (1996): 884-888.

Shiohara T, Yoshizumi M, Takagi Y, Izumi T 2013. Future prospects of high Tc

superconductors-coated conductors and their applications, Physica C: Superconductivity 484,

15 January 2013: 1-5.

Sinclair G, Klepper S, and Cohen W 2000. What’s experience got to do with it? Sources of cost

reduction in a large specialty chemicals producer. Management Science, 46: 28–45.

Smith R 1988. A Historical Overview of Computer Architecture, IEEE Annals of the History of

Computing 10(4): 277-303.

SingularityHub.com, 2013. http://singularityhub.com/2012/09/17/new-software-makes

-synthesizing-dna-as-easy-as-using-an-ipad/, last accessed on November 27, 2013.

Solow R. 1957. Technical Change and the Aggregate Production Function, Review of Economics

and Statistics, 39: 312-320.

Stasiak J, Richards S, and Angelos S 2009. Hewlett Packard's Inkjet MEMS Technology, Proc.

of Society of SPIE: 7318, http://144.206.159.178/ft/CONF/16431771/16431793.pdf

Stobaugh, R 1988. Innovation and Competition: The Global Management of Petrochemical

Products. Boston: Harvard Business School Press

Suzuki T, 2010, Challenges of Image-Sensor Development, Int’l Solid State Sensors Conference

Tarascon, J. 2009. Batteries for Transportation Now and In the Future, presented at Energy 2050,

Stockholm, Sweden, October 19-20.

Thompson P 2012. The Relationship between Unit Cost and Cumulative Quantity and the

Evidence for Organizational Learning-by-Doing, Journal of Economic Perspectives 26(3):

203-224.

Thornton R and Thompson P 2001. Learning from Experience and Learning from Others,

American Economic Review 91(5): 1350-1368.

Utterback, J., 1994, Mastering the dynamics of innovation, Harvard Business School Press.

Winter S 2008. Scaling heuristics shape technology! Should economic theory take notice?

Industrial and Corporate Change 17(3): 513–531.

Wright T P, 1936. Factors Affecting the Cost of Airplane, Journal of Aeronautical Sciences,

3(4): 122 – 128.

Table 1. Technologies with Recent Rapid Rates of Improvement

Technology

Domain

Sub-Technology Dimensions of measure Time

Period

Improvement

Rate Per

Year

Energy

Trans-

formation

Light Emitting

Diodes (LEDs)

Lumens per Dollar, white 2000-2010 40.5%

Organic LEDs Luminosity per Watt,

green

1987-2005 29%

GaAs Lasers Power density 1987-2007 30%

Cost per Watt 1987-2007 31%

Quantum Dot

Displays, Solar Cells

Efficiency, red displays 1998-2009 36.0%

Efficiency of Solar Cells 2010-2013 42.1%

Energy

Trans-

mission

Superconductors:

BSSCO and YBCO

Current-length per cost 2004-2010 115%

Current x length - BSSCO 1987-2008 32.5%

Current x length - YBCO 2002-2011 53.3%

Information

Trans-

formation

Microprocessor ICs Number of transistors/chip 1971-2011 38%

Camera chips Pixels per dollar 1983-2013 48.7%

MEMS: Artificial Eye Number of Electrodes 2002-2013 45.6%

MEMS: inkjet

printers

Drops per second 1985-2009 61%

Organic Transistors Mobility 1982-2006 109%

Single Walled Carbon

Nanotube Transistors

1/Purity (% metallic) 1999-2011 32.1%

Density 2006-2011 357%

Superconducting

Josephson Junctions

Qubit Lifetimes 1999-2012 142%

Bits per Qubit lifetime 2005-2013 137%

Photonics Data Capacity per Chip 1983-2011 39.0%

Table 1. Technologies with Recent Rapid Rates of Improvement (continued)

Technology

Domain

Sub-Technology Dimensions of measure Time

Period

Improvement

Rate Per

Year

Information

Trans-

formation

Computers Instructions per unit time 1979-2009 35.9%

Instructions per time-cost 1979-2009 52.2%

Quantum Computers Number of Qubits 2002-2012 107%

Information

Storage

Flash Memory Storage Capacity 2001-2013 46.8%

Resistive RAM Storage Capacity 2006-2013 272%

Ferroelectric RAM Storage Capacity 2001-2009 37.8%

Magneto RAM Storage Capacity 2002-2011 57.8%

Phase Change RAM Storage Capacity 2004-2012 63.1%

Magnetic Storage Recording density of disks 1991-2011 55.7%

Recording density of tape 1993-2011 32.1%

Cost per bit of disks 1956-2007 32.7%

Information

Transmission

Last Mile Wireline Bits per second 1982-2010 48.7%

Wireless, Cellular Bits per second 1996-2013 79.1%

Wireless, WLAN 1995-2010 58.4%

Wireless, 1 meter 1996-2008 77.8%

Biological

Trans-

formation

DNA Sequencing per unit cost 2001-2013 146%

Synthesizing per unit cost 2002-2010 84.3%

(Azevedo et al, 2009; Haitz and Tsao, 2011; Lee, 2005; Sheats et al, 1996; Martinson, 2007;

Kwak, 2010; NREL, 2013.Shiohara et al, 2013; Selvamanickam V 2011; Wikipedia, 2014; Preil,

2012; Chader, 2009; Stasiak et al, 2009; Hasegawa and Takeya, 2009; Franklin, 2013; Devoret

and Schoeldopf, 2013; Evans et al, 2011; Koomey et al, 2011; D-Wave, 2013; ISSCC, 2013;

Francis, 2011; Yoon, 2010; Brown, 2011; ISSCC, 2013; NHGRI, 2013; SingularityHub.com,

2013)

0

5

10

15

20

25

-10 -8 -6 -4 -2 0 2 4 6 8

10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 >42

FIgure 1. Number of Technologies by Annual Rates of Improvement

Annual Rates of Improvement

0

2

4

6

8

10

12

14

0 2 4 6 8 10 12

FIgure 2. Number of Chemical Technologies by Annual Rates of Improvement

Annual Rates of Improvement

0

5

10

15

20

-10 -8 -6 -4 -2 0 2 4 6 8 10 12 14

Figure 3. Number of Non-Chemical Technologies by Annual Rates of Improvement

Annual Rates of Improvement

0

2

4

6

8

25 35 45 55 65 75 85 95 105 115 125 135 145 200 300

Figure 4. Number of Technologies with Annual Rates of Improvement Greater than the Annual Rates Shown on the X-Axis

Annual Rates of Improvement