University Software Ownership and Litigation: A First Examination

Abstract

Software patents and university-owned patents represent two of the most controversial intellectual property developments of the last twenty-five years. Despite this reality, and concerns that universities act as “patent trolls” when they assert software patents in litigation against successful commercializers, no scholar has systematically examined the ownership and litigation of university software patents. In this Article, we present the first such examination. Our empirical research reveals that software patents represent a significant and growing proportion of university patent holdings. Additionally, the most important determinant of the number of software patents a university owns is not its research and development (“R&D”) expenditures (whether computer science-related or otherwise) but, rather, its tendency to seek patents in other areas. In other words, universities appear to take a “one size fits all” approach to patenting their inventions. This one size fits all approach is problematic given the empirical evidence that software is likely to follow a different commercialization path than other types of invention. Thus, it is perhaps not surprising that we see a number of lawsuits in which university software patents have been used not for purposes of fostering commercialization, but instead, to extract rents in apparent holdup litigation. The Article concludes by examining whether this trend is likely to continue in the future, particularly given a 2006 Supreme Court decision that appears to diminish the holdup threat by recognizing the possibility of liability rules in patent suits, as well as recent case law that may call into question certain types of software patents.

Introduction

Software patents and university-owned patents represent two of the most controversial intellectual property developments of the past few decades. Various scholars have quarreled with the alleged vagueness and undue breadth of software patent claims.¹ Some have software processes that are not limited to a physical transformation or machine.⁶ In contrast with software patents, university-owned patents have existed for more than a few decades. The number of university-held patents has increased substantially, however, since the 1980 passage of the Bayh-Dole Act.⁷ While the legal question was sometimes murky prior to 1980, Bayh-Dole made it unequivocally clear that universities can patent federally-funded research.⁸Assertive university patenting has attracted attention in both scholarly and popular literature.⁹ Additionally, because universities, and sometimes even their exclusive licensees, are nonmanufacturing patentees, the intense debate over whether such patentees employ “holdup” strategies deleterious to innovation when they assert patentsagainst successful commercializers directly implicates universities.¹⁰ That said, almost all analyses of university patenting have focused on patenting within the life sciences.¹¹ This focus is perhaps not surprising, as the major economic argument put forward in the legislative history of the Bayh-Dole Act—that patents on publicly-funded invention promote commercialization of such invention¹²—would appear to apply most clearly to areas of the life sciences such as drug development. Conventional wisdom suggests that, without the quasi-monopoly protection of a patent on the small molecule chemical, few firms would be interested in taking a potentially promising drug candidate through the expensive clinical trial and approval process.¹³

In the case of publicly-funded software, by contrast, the need for a patent and exclusive license to promote further development is less apparent.¹⁴ Although development costs are not uniformly low, they are likely to be low relative to those in the biopharmaceutical industry.¹⁵ Indeed, in certain cases of open source software development, firms derive revenue not from property rights over the software product itself, but from a strategy that monetizes the value of support services and complementary hardware.¹⁶ Thus, if universities are in fact making strong proprietary claims on software, scholars should be concerned. The social welfare argument for such claims is more tenuous than in the life sciences.

Despite IT sector complaints about university behavior,¹⁷ as well as prominent lawsuits involving software patents,¹⁸ the subject of university software ownership and litigation has not been studied systematically. Indeed, this Article represents the first systematic study of which we are aware.¹⁹ We rely in part on a unique, handcurated database of university software patents.²⁰ This quantitative analysis is supplemented by interviews conducted with technology transfer officers at those universities that own large numbers of software patents, as well as academic scientists and other officials prominent in the open source movement. The combination of our quantitative and qualitative inquiry yields a number of important results. First, software patents represent a significant and growing percentage of university patent holdings.²¹ Second, university software patenting practices tend to mimic their nonsoftware patenting practices. The data suggests that those universities that have a higher patent propensity in general are also more likely to obtain software patents.²² Similarly, our interviews show that some universities view software as similar to other, more physical inventions.²³ The difficulty with this view is that software is likely to follow a different commercialization path than other inventions. Thus, it is perhaps not surprising that we see a fair number of litigated cases involving software patents, and that almost all of these appear to represent situations where the university and/or its exclusive licensee is asserting the patent against an entity that has successfully commercialized software independent of the patent.²⁴

Notably, in the majority of these cases, the university's argument has lost on grounds of either patent invalidity or noninfringement.²⁵

The main rationale for supporting patenting would, therefore, appear to be the promotion of start-up businesses, presumably on the theory that start-ups, and market-based activities more generally, are likely to be more innovative than activities in large, vertically-integrated incumbent firms.²⁶ However, whether patents are necessary for start-up promotion is not as clear in the software area as it is, for example, in the biotechnology industry. Moreover, in contrast with biotechnology, where copyright is not available, universities can use software copyright to achieve revenue generation goals.²⁷

It may be that university software patenting is a transient phenomenon. As we discuss below,²⁸ some technology transfer officers argue that they no longer view the patenting of software as a particularly good mechanism for technology transfer.²⁹ In addition, the 2006 Supreme Court decision in eBay Inc. v. MercExchange, L.L.C.³⁰ allowing district courts discretion to award damages even after validity and infringement have been found, coupled with recent Federal Circuit case law that may call into question certain types of software patents, could make software patenting and litigation less attractive to universities in the future.

This Article proceeds as follows. Part I discusses the history of software patenting and how it influenced our methodology for compiling a database of university software patents. Next, Part II presents our quantitative results; specifically, we identify trends in, and determinants of, university software patenting. Part III integrates these quantitative results with results from our interviews. Part IV then discusses the “holdup” features present in much university litigation over software patents. More generally, it provides a policy analysis of why universities' apparent “one size fits all” approach to patenting is problematic. Finally, Part V discusses the path forward, with a focus on whether we are likely to see more of this unitary approach (and concomitant litigation) in the future.

I. Software Patenting: History and Methods of Identifying

To identify university patents, we began by using the U.S. Patent and Trademark Office's (“PTO”) Cassis database to identify all patents issued in 1982, 1987, 1992, 1997, and 2002 that were assigned to institutions classified as Research or Doctoral Universities in the Carnegie Commission of Higher Education's 1972 or 1994 reports. We chose these particular years because they span a series of shifts in the legal regime surrounding software produced at universities.³¹ Not only did university patenting increase significantly after the 1980 passage of the Bayh-Dole Act,³² but patent jurisprudence in the area of software also evolved considerably. Because this evolution is closely related to the manner in which we define the term “software patent,” we describe it in some detail below.

A. The History of Software Patents

In the 1970s, the dominant intellectual property regime for software was copyright, not patent.³³ A 1972 Supreme Court case, Gottschalk v. Benson,³⁴ appeared to reject software (in that case a computerized method for converting decimal numbers to binary numbers) as patentable subject matter on the grounds that patent law did not encompass abstract scientific or mathematical principles.³⁵ Further, several years later in the 1980 amendments to the Copyright Act of 1976, Congress expressly endorsed copyright as an appropriate protection regime for software.³⁶

The intellectual property terrain shifted in the 1980s. In the 1981 decision Diamond v. Diehr,³⁷ the Supreme Court gave its first clear indication that certain types of software-implemented inventions were patentable. Diehr narrowed Gottschalk by upholding as patentable subject matter a rubber-curing process that used software to calculate cure time. According to the Diehr Court, the physical transformation of the rubber “into a different state or thing” took the invention being claimed out of the realm of abstraction.³⁸

Through the 1980s, the Court of Appeals for the Federal Circuit generally followed a test somewhat similar to that enunciated in Diehr. Under this test, if an invention's claims involved nothing more than an algorithm, then the invention could not be patented.³⁹ However, if the claims involved a mathematical algorithm that was “applied to, or limited by, physical elements or process steps,” such claims would constitute patentable subject matter.⁴⁰ The overall message to patent attorneys was that software could be patented, but it had to be claimed as something else. Assuming an approximate examination pendency of two years during the relevant period,⁴¹ patents issued in our sample years of 1987 and 1992 should reflect patenting of software as “something else.”

Moreover, as the patent option was becoming more attractive, copyright was becoming much less so. In the early 1990s, a series of appellate court decisions made it clear that copyright covered primarily the literal source code of the program.⁴²

Greater changes lay in store. In the 1994 case of In re Alappat,⁴³ the Federal Circuit effectively eliminated any limitation on patenting software by concluding that subject matter criteria could be met by claiming software as a new machine—a “special purpose” computer—when it was executed.⁴⁴ Presumably all software could produce such a special purpose computer and hence be patentable. Again, assuming a two-year examination pendency, patents issued in 1997 should begin to incorporate any effects that decisions like Alappat had on filing incentives. Also, in 1996, the PTO issued software guidelines that broadly allowed software as patentable subject matter whether the software was claimed as a machine or a process.⁴⁵ Two years later, the Federal Circuit's decision in State Street v. Signature Financial Group⁴⁶ similarly rejected any special subject matter test for software, finding that software (like all inventions) is patentable if it produces a “useful” result.⁴⁷ After State Street, there was no need for even the fig leaf of a physical machine or process. Thus, patents issued in 2002 should incorporate any impact of the PTO guidelines and the State Street decision on filing incentives.⁴⁸

B. Identifying University Software Patents

Given this history, it is perhaps not surprising that identifying “software” patents is very difficult. Even now, there is no universally accepted definition of what a software patent is. To our knowledge, there have been only a few significant efforts to identify a large data set of software patents.⁴⁹ An initial paper by Stuart Graham and David Mowery,⁵⁰ which does not attempt to define the term “software patent,” relies upon certain International Patent Classifications (“IPCs”)⁵¹ as limited to patents in those classes owned by large software firms.⁵² A more recent paper by Graham and Mowery uses particular U.S. patent classes, once again limited to patents in those classes which are owned by large software firms.⁵³ The Graham and Mowery approach is not likely to be significantly overinclusive so long as it is limited to patents owned by packaged software firms. However, their approach may be quite underinclusive, missing software patents assigned to other firms as well as software patents in other patent classes.⁵⁴

Another significant effort to identify a large set of software patents, by James Bessen and Bob Hunt,⁵⁵ defines “software patent” to include patents on inventions in which the data processing algorithms are carried out by code either stored on a magnetic storage medium or embedded in chips (“firmware”).⁵⁶ Rejecting the use of patent classifications,⁵⁷ Bessen and Hunt study a random sample of patents and classify them according to their definition. Using characteristics of patents Bessen and Hunt find to fit their definition, they then develop a keyword search algorithm to identify software patents.⁵⁸

Although the Bessen and Hunt definition of software patent is reasonable, there are pitfalls associated with using automated keyword searches to identify such patents. From one of the co-author's (John Allison's) study of thousands of computer-related patents, it is clear that the use of language in the titles, abstracts, written descriptions, and claims of patents, even in those dealing with the same area of technology, can be highly idiosyncratic among different patent owners. In general, a common criticism of patents in the IT industries is that the industries lack a standardized vocabulary for claims.⁵⁹ Moreover, software is a critical part of inventions in so many fields that reliance on particular search terms could produce a data set that has both false positives and false negatives.⁶⁰

Instead of relying upon prior approaches, we formulated a definition of “software patent” based on the extensive experience one of us (Allison) has had in reading software patents. Our definition of software patent is “a patent in which at least one claim element consists of data processing—the actual manipulation of data—regardless of whether the code carrying out that data processing is on a magnetic storage medium or embedded in a chip.” Not only is it possible to apply the definition consistently, but it also captures the realities of claim drafting. In Allison's experience, patent claims often include the prior art, with only one or two elements covering the purportedly novel and nonobvious advance. For example, a claim may read as though it covers a generic router, printer, magnetic resonance imaging machine, or other hardware, when in fact the only purported novelty is in one element consisting of a function carried out by algorithms. Some of this may be a consequence of the fact that, prior to Alappat and the 1996 PTO guidelines, software had to be claimed not as a new algorithm per se but instead as a new machine that allegedly did something different because of the new algorithm.⁶¹ More fundamentally, a claim covers the entire invention, and in a case like this, the entire invention is not just the new algorithm in isolation; rather, it is a piece of hardware that allegedly does something different because of the new algorithms.

In a large set of patents, it is a practical impossibility to include only those patents in which the software element (as contrasted with other elements of the invention) is novel and nonobvious. In order to restrict ourselves to patents in which the software element was novel and nonobvious, we would need to employ a person having ordinary skill in the art to conduct a very thorough study of the relevant prior art. Even then, the question would be plagued by doubt because issues of novelty and nonobviousness are typically difficult to resolve even after a full evidentiary exploration in court. But the fact that, under our definition, the data processing must be identified in a claim element does suggest that software is sufficiently important to novelty and nonbviousness for the patent claim drafter to include it as a limitation of the claim (thereby narrowing the claim). In other words, it would be foolhardy for a patent attorney to include such an element unless it was a critical part of the invention.

Our approach does have limits. One problem with our approach is that it involves the slow and laborious process of reading patents. Although the decision on many patents is clear, there will always be a substantial percentage that must be studied with great care.⁶² Moreover, a degree of subjective judgment is occasionally required. However, at least in the case of a relatively small data set, we believe that increased accuracy more than compensates for time intensity and the absence of algorithmic criteria readily replicable by automated methods. We do not claim that our data set of university-owned software patents is perfect, but we do contend that our error rate is very small—certainly smaller than in any data set acquired by means of patent classifications or keyword searches.

In addition to identifying which patents out of the more than 7,600 university-owned patents in our sample are software patents, we also identified a subset of those that may be called “pure software patents.” These are patents in which the claims consist only of data processing—that is, the entire invention consists of algorithms.⁶³ This task required thorough study of each of the patents that had already been identified as a software patent. Although the process of identifying “pure” software patents was accomplished with a high degree of accuracy, there was a small number about which reasonable minds could differ. Thus, for this second stage, we also do not profess to have achieved perfection, but we do maintain that our error rate is very low.

II. University Software Patenting: Trends, Determinants, and Department Origin

A. Trends

Figure 1 shows that university software patenting increased more than ten-fold over the 1982–2002 period, from thirty-seven patents in 1982 to 396 patents in 2002. Over this period, the “pure software” proportion of university software patents also increased dramatically, from thirteen percent to thirty-two percent of all university software patents. The latter change is hardly surprising. While the patentability of pure software was unclear in the 1980s, its status became much more secure in the 1990s.⁶⁴ Over these two decades, university software patenting also grew at a faster rate than university patenting overall. As a consequence, the software share of university patents rose from nine percent in 1982 to thirteen percent in 2002, as seen in Figure 2.

Figure 1. University Software Patents, By Type and Year.

Figure 2. University Software Patents as a Share of All University Patents.

Table 1 below lists the fifteen universities that received the most software patents in 2002.⁶⁵ Together, these fifteen institutions accounted for sixty percent of all university software patents issued in 2002. The top five institutions alone, the Massachusetts Institute of Technology (“MIT”), the University of California, Stanford University (“Stanford”), the California Institute of Technology (“Caltech”), and the University of Texas, accounted for over a third (34.2 percent) of all university software patents. The top five patentees also represented the top five university patentees overall in 2002. However, moving further down the list of the top fifteen, we see that a number of the top software patentees are not among the top university patentees overall. The University of Washington (sixth in software patenting/fifteenth in overall patenting), the Georgia Institute of Technology (“Georgia Tech”) (eighth/twentieth), Carnegie Mellon (ninth/fifty-first), the University of Rochester (twelfth/fiftieth) and the University of Illinois (fourteenth/twenty-eighth) particularly stand out as institutions substantially more prominent in software patenting than overall patenting.

Table 1. University Software Patenting, Overall Patenting, and Pure Software Patenting in 2002.

	Rank in Patenting (Issue Year 2002)
University	Software	Overall	Pure Software
Massachusetts Institute of Technology	1	2	1
University of California	2	1	2
Stanford University	3	4	4
California Institute of Technology	4	3	23
University of Texas	5	5	41
University of Washington	6	15	6
University of Wisconsin	7	7	8
Georgia Institute of Technology	8	20	3
Carnegie Mellon University	9	51	13
Johns Hopkins University	10	6	11
State University of New York	11	8	20
University of Rochester	12	50	14
University of Pennsylvania	13	13	42
University of Illinois	14	28	5
Columbia University in the City of New York	15	14	10

With respect to the patenting of “pure” software, Table 1 shows that the top three patenting institutions overall also ranked among the top recipients of pure software patents (first, second, and fourth). In contrast, although Caltech and the University of Texas ranked high in overall software patenting (fourth and fifth, respectively), they ranked relatively low in the patenting of pure software (twenty-third and forty-first, respectively). The University of Washington, Georgia Tech, Carnegie Mellon, the University of Rochester, and the University of Illinois—mentioned earlier as standing out in software patenting relative to overall patenting—also stand out with respect to numbers of pure software patents (sixth, third, thirteenth, fourteenth, and eleventh, respectively).

As these examples suggest, several factors may affect university software patenting. First, the amount of software-related R&D, and thus the output of software, may matter. Second, the size of overall research enterprise may matter, because software can be developed in many parts of the university. Third, an individual university's nonsoftware-related propensity to patent, i.e., the share of nonsoftware research outputs it patents (either because its researchers are prone to file invention disclosure statements and push for patents on those disclosures or because technology transfer officers are prone to seek patents on disclosures), may also affect software patenting.⁶⁶ Although there are undoubtedly other factors that affect software patenting—for example, how software ownership in particular is viewed in the technology licensing office or among faculty⁶⁷—these factors are difficult to measure quantitatively. We examine some of these factors in our qualitative analyses below.

B. Patent Production Functions

To examine the relationship of software R&D, overall R&D, and nonsoftware patent propensity to university level software patenting, we collected data on R&D funds as well as nonsoftware patenting for all of the 202 Carnegie research universities in our sample. The unit of analysis for the “patent production functions” we estimate is a university in a given issue year. For each institution, we created variables measuring the number of software patents and number of other patents issued in 1982, 1987, 1992, 1997, and 2002. We used the National Science Foundation's WebCaspar database⁶⁸ to collect R&D data for these universities for the 1977–2002 time period.⁶⁹ Specifically, we collected data on computer science R&D and overall R&D and used this information to construct variables measuring the sum of computer science and other R&D over the previous five years. The main empirical analyses relate a university's software patenting to its nonsoftware patenting in a given issue year and to its computer science R&D and other R&D over the five years prior. We decided to use a five-year time frame because it is difficult to estimate precisely when the research that led to a particular invention disclosure (let alone patent application or issued patent) was conducted. By using a relatively long time frame, we aimed to get a sense of the research commitment of the university, both inside and outside the software arena, around the time an invention disclosure was likely to have been submitted.⁷⁰

To facilitate interpretation, we took natural logarithms of the independent variables.⁷¹ Because the dependent variables are integer valued, we estimated negative binomial regressions relating software patents and pure software patents to research expenditures.⁷² Tables 2 through 6 show the main results from these simple cross-sectional regressions for 1982 to 2002, respectively. The dependent variable in Models 1 and 2 is the number of software patents issued to a university in a given year. In Models 3 and 4, the dependent variable is the number of pure software patents.

Table 2. Negative Binomial Models, Determinants of 1982 Software Patenting.

	Model1	Model2	Model3	Model4
	(1)	(2)	(3)	(4)
In Lagged CS Funds	.218* (.128)	.167 (.124)	.770** (.336)	.748** (.358)
In Lagged Other Funds	.612** (.270)	−.043 (.193)	1.435** (.635)	.709 (.774)
In Other Patents		1.326*** (.272)		.742 (.501)
Const.	−10.317*** (2.824)	−4.045** (1.798)	−27.910*** (8.697)	−19.811** (10.044)
Obs.	202	202	202	202

Robust standard errors in parentheses.

^***denotes p<.001,

^**denotes p<01, and

^*denotes p<.05.

Table 6. Negative Binomial Models, Determinants of 2002 Software Patenting.

	Model1	Model2	Model3	Model4
	(1)	(2)	(3)	(4)
In Lagged CS Funds	.110** (.049)	.097** (.045)	.588*** (.121)	.523*** (.110)
In Lagged Other Funds	.839*** (.118)	.073 (.141)	.223 (.161)	−.395** (.163)
In Other Patents		.910*** (.141)		.781*** (.167)
Const.	−11.577*** (1.406)	−3.613** (1.510)	−9.017*** (1.615)	−2.182 (1.576)
Obs.	202	202	202	202

Robust standard errors in parentheses.

^***denotes p<.001,

^**denotes p<.01, and

^*denotes p<.05.

In negative binomial models, coefficients on log-transformed variables can be interpreted as elasticities. Model 1 for the 1982 sample (Table 2) shows that both computer science R&D and other R&D are related to software patenting. Upon introducing the nonsoftware patenting variable, neither of the R&D variables is significant, but the number of nonsoftware patents is, with a one percent increase in other patenting associated with a 1.3 percent increase in software patenting. Both software and nonsoftware R&D are related to the number of pure software patents (Model 3), but the latter is no longer statistically significant after introducing the nonsoftware patenting variable. Thus, the only clear implication of the 1982 cohort is the strong correlation of nonsoftware patenting with overall software patents.

Tables 3 and 4 show that, in 1987 and 1992, once nonsoftware patents are taken into account, neither computer science nor other R&D has a consistently strong association with either overall software patenting or pure software patenting. Nonsoftware patenting is, however, strongly related to software patenting and pure software patenting, after controlling for R&D. Depending on the specific model and gear, elasticities range from 0.797 to 1.035.

Table 3. Negative Binomial Models, Determinants of 1987 Software Patenting.

	Model1	Model2	Model3	Model4
	(1)	(2)	(3)	(4)
In Lagged CS Funds	.343*** (.122)	.314*** (.092)	.289 (194)	.210 (.162)
In Lagged Other Funds	1.013*** (.236)	.291 (.198)	.864** (.351)	.094 (.319)
In Other Patents		.797*** (.163)		1.035*** (.303)
Const.	−16.411*** (2.592)	−8.430*** (2.117)	−15.243*** (3.792)	−6.796** (3.261)
Obs.	202	202	202	202

Robust standard errors in parentheses.

^***denotes p<.001,

^**denotes p<.01, and

^*denotes p<.05.

Table 4. Negative Binomial Models, Determinants of 1992 Software Patenting.

	Model1	Model2	Model3	Model4
	(1)	(2)	(3)	(4)
In Lagged CS Funds	.143* (.077)	.082 (.073)	.122 (.110)	.071 (.107)
In Lagged Other Funds	1.083*** (.192)	.348* (.206)	.968*** (.287)	.264 (.323)
In Other Patents		1.027*** (.199)		.955*** (.318)
Const.	−15.295*** (2.213)	−7.610*** (2.221)	−14.735*** (3.357)	−7.392** (3.446)
Obs.	202	202	202	202

Robust standard errors in parentheses.

^***denotes p<.001,

^**denotes p<.01, and

^*denotes p<.05.

Further, Tables 5 and 6 show that after the mid-1990s—the post-Alappat⁷³ era—universities that had larger stocks of computer science R&D also tended to do more software patenting and pure software patenting. This conclusion is consistent with the descriptive data for issue year 2002 in Table 1. In each of the models, however, after controlling for computer science R&D and other R&D, nonsoftware patenting was significantly related to both software patenting and pure software patenting, with elasticities ranging from 0.78 to 0.91 depending on the specific model and year.

Table 5. Negative Binomial Models, Determinants of 1997 Software Patenting.

	Model1	Model2	Model3	Model4
	(1)	(2)	(3)	(4)
In Lagged CS Funds	.245*** (.076)	.172** (.072)	.348*** (.134)	.275** (.131)
In Lagged Other Funds	.986*** (.160)	.308 (.196)	1.005*** (.252)	.271 (.320)
In Other Patents		.832 *** (.185)		.813*** (.282)
Const.	−14.941*** (1.878)	−7.394*** (2.179)	−17.358*** (2.975)	−9.023** (3.587)
Obs.	202	202	202	202

Robust standard errors in parentheses.

^***denotes p<.001,

^**denotes p<.01, and

^*denotes p<.05.

The finding that nonsoftware patenting (conditional on overall and computer science R&D) has a qualitatively and statistically significant relationship with software patenting suggests that universities with a higher nonsoftware patenting propensity also tend to patent more software.

This correlation between nonsoftware and software patenting propensity could be driven by a number of factors. One potentially omitted variable from the analyses above is university success in getting patent applications issued. If some universities are systematically better at securing patents, this could drive the observed correlation even if the propensity to file patent applications on software is unrelated to the propensity to file applications on other inventions. We explored this question for the 2002 cohort for the subset of 130 universities where AUTM data on “application success” were available.⁷⁴

As Models 1 and 2 in Table 7 show, for all software patents and pure software patents respectively, application success is not significantly related to software patenting.

Table 7. Negative Binomial Models, Determinants of 2002 Software Patenting, AUTM Universities.

	Model1	Model2	Model3	Model4
	(1)	(2)	(3)	(4)
In Lagged CS Funds	.100** (.046)	.537*** (.111)	.091 (.057)	.583*** (.120)
In Lagged Other Funds	−.040 (.169)	−.524** (.235)	.428** (.202)	−.008 (.254)
In Other Patents	.807*** (.160)	.717*** (.219)
Application Success Rate	−.207 (.360)	.068 (.426)
In TTO FTEs			.503** (208)	.407* (.247)
Const.	−1.663 (1.914)	−.317 (2.530)	−6.420*** (2.468)	−6.356** (3.046)
Obs.	130	130	101	101

Robust standard errors in parentheses.

^***denotes p<.001,

^**denotes p<.01, and

^*denotes p<.05.

Another possible explanation for the correlation between nonsoftware and software patenting propensity could be scale economies: the marginal costs of obtaining a software patent (or any additional patent) may be lower for universities that have larger technology transfer operations. To test this possibility, we used data on the size of technology transfer offices. Reports from the Association of University Technology Managers contain data on the number of licensing officers employed by each university in 2002, though only for a subset (102) of universities in our sample.⁷⁵ Models 3 and 4 in Table 7 show that, after controlling for nonsoftware and software R&D, this variable is significantly related to both pure and software patenting, consistent with the scale economies hypothesis.

Finally, we estimated a panel version of these regressions for the entire period with university and year fixed effects. In this model, the estimated changes in R&D and other patenting are identified using within-university variation over time. Thus, any time invariant university factors (including “fixed” university technology transfer abilities) would not affect estimates from this model.

Table 8 shows the results. Notably, the year dummies are positive and their magnitude increases over time, reflecting the overall growth in university patenting over the 1982–2002 period. Model 1 shows that the main factor affecting overall software patenting is changes in nonsoftware patenting, with a one percent increase in nonsoftware patenting implying a .46 percent increase in software patenting. The corresponding finding for pure software patenting is neither qualitatively nor statistically significant. Additionally, none of the R&D measures are statistically significant for either software or pure software. This last set of results should be interpreted with caution, however. Because the panels are short and there is limited within-university variation, it is difficult to draw strong inferences.

Table 8. Negative Binomial Models, Determinants of Software Patenting, Panel Model.

	Model1	Model2
	(1)	(2)
Log Lagged CS Funds	.039 (.047)	.107 (.096)
Log Lagged Other Funds	−.387 (.289)	−.118 (.628)
Log Other Patents	.457*** (.093)	.054 (.167)
Year Dummy 1987	.688*** (.244)	1.363* (.583)
Year Dummy 1992	1.381*** (.324)	2.193*** (.754)
Year Dummy 1997	1.767*** (.400)	2.708*** (.897)
Year Dummy 2002	2.090*** (.472)	3.045*** (1.040)
Const.	−17.388 (5379.716)	−15.120 (418.959)
Obs.	1010	1010

All models include university fixed effects. The Left-out year category is 1982. Robust standard errors in parentheses.

^***denotes p<.001,

^**denotes p<.01, and

^*denotes p<.05.

Taken together, we read these results as consistent with the impressions from our interviews (discussed below) that university software patenting practices are related to their nonsoftware patenting practices. An ideal test of this hypothesis would control for the number of disclosed software inventions for which patents would indeed facilitate technology transfer. If we had such a measure, and found that nonsoftware patent propensity nonetheless affected the volume of software patenting, this would provide strong support for the argument that university software patenting is less about “technology transfer” and more about “business as usual.” Unfortunately, such a measure is unlikely to be obtained.⁷⁶

C. Departmental Origin

To explore the sources of software patenting in greater detail, we also identified the departmental origin of the inventors in the top fifteen academic software patentees in 2002. As noted above, these institutions accounted for sixty percent of 2002′s overall academic software patenting.⁷⁷

These 241 patents had 544 distinct inventors. Based on web searches, we were able to locate the primary departmental affiliation for 73.5 percent (400) of these inventors. Seen another way, these 241 patents include 661 unique inventor-patent dyads. For example, if an inventor is included on three patents, she generates three inventor-patent dyads. For 26.5 percent of these dyads, we could not identify the inventor's department. EE/CS departments. Moreover, consistent with arguments that software is produced across the university, a number of biomedical departments are also represented, including Neuroscience, Radiology, and Medical Physics. To the extent that software is being treated just like other inventions for purposes of patenting, this may be in part because a significant proportion of it emerges from biomedical departments where patenting is the norm. In this regard, it bears emphasis that, of the 970 software patents in our overall sampling, 333 (more than one-third) represented software with biomedical applications.

Table 10 shows the analogous table for pure software patents. Perhaps not surprisingly, more than half of the patent-inventor dyads on pure software patents emanate from electrical engineering, computer science or joint EE/CS departments. Other departments are less prominent in production of pure software patents than they are in the production of other software patents.

Table 10. Departmental Origin of Inventors for Pure Software Patents Issued to the Top Fifteen Recipients of Software Patents in Issue Year 2002.

Department	N	Percent	Cumulative Percent
Computer Science	33	23.57	23,57
EE/CS	29	20.71	44.29
Electrical Engineering	12	8.57	52,86
Lincoln Laboratory	6	4.29	57.14
Aeronautics and Astronautics	5	3.57	60.71
Robotics Institute	5	3.57	64.29
Agricultural and Biological Engineering	3	2.14	66.43
Chemistry and Biochemistry	3	2.14	68.57
Mathematics	3	2.14	70.71
Bioslatislics and Medical Informatics	2	1.43	72.14

Counts and percentages calculated at the patent-invenlor level

III. University Ownership Policies

Our quantitative results demonstrate that, on the whole, for software patent applications filed in the 1980s and 1990s, universities appeared to take a “one size fits all” approach to technology. Those universities that tended to file a lot of nonsoftware patents also filed a lot of software patents (both ordinary software and pure software). The regression results underscore what can be seen through a casual review of the descriptive statistics in Table 4: the top five software patentees in 2002—MIT, University of California, Stanford, Caltech, and the University of Texas—were also the top five overall patentees. Notably, none of these five was in the top five in computer science R&D spending for the years 1996–1998.

On the other hand, a qualitative review of our descriptive data also indicates some anomalies in this general pattern. For example, of the top five software patentees in 2002, two—Caltech and the University of Texas—did not have significant numbers of pure software patents.⁷⁸ Moreover, five universities—the University of Washington, Georgia Tech, Carnegie Mellon University, the University of Rochester, and the University of Illinois—rank substantially higher in software/pure software patenting than in overall patenting.

In order to get a more fine-grained sense of factors that might be influencing individual universities, we investigated technology transfer policies at the fifteen universities that received the most software patents in 2002. In addition to reviewing publicly- available materials, we conducted interviews with two groups of relevant parties: (1) technology transfer managers responsible for software at the universities;⁷⁹ and (2) academics and graduate students prominent in software development. The latter group was selected through snowball sampling. Interviews were conducted by telephone, lasted approximately forty-five minutes to one hour, and were semi-structured. On occasion, follow-up discussions were necessary and were conducted either via e-mail or by telephone. A list of questions asked in each interview is attached as Appendix A.⁸⁰

As an initial matter, several of the interviews confirmed the results of our quantitative analysis—that technology transfer officers treat software like other inventions. Lita Nelsen, director of the MIT technology transfer office noted that, when discussing pure software, “if there are no strong feelings on the part of the authors to open source their work, we will look at it like any other invention.”⁸¹

Our qualitative investigation also suggested reasons for the relatively low numbers of pure software patents at the University of Texas and Caltech. At the University of Texas, complaints from computer science faculty resulted in the explicit adoption of a policy allowing faculty freedom to share their work as they wanted.⁸² Such freedom includes express permission to use the GNU General Public License (“GPL”), a “viral” open source license that requires software source code to be open but also requires those who redistribute the software to make any modifications they have made to the source code available.⁸³ This policy was adopted in the 1990s and, therefore, would have affected the number of pure software patents that issued in 2002.⁸⁴

The explanation for Caltech's small number of pure software patents is less transparent. Because we were unable to speak with the technology licensing office,⁸⁵ our knowledge of the Caltech situation is based on reports from scientists who work there. According to one prominent open source software developer at Caltech, most software developed there is for internal, in-house use and is probably not even reported to the technology transfer office.⁸⁶ For software that is reported, the faculty researcher decides how the software should be exploited.⁸⁷ The technology transfer office is not only familiar with open source, but is apparently eager to use the viral GPL license; the viral GPL allows for the future possibility of “forking,” with one fork continuing to be available without charge via the GPL and the other fork converted into a nonviral license for which corporate clients might be willing to pay.⁸

In contrast, the positions of two of the other top five software patentees, MIT and the University of California, are (or at least have been) less explicitly favorable to open source.⁸⁹ According to MIT's director of technology transfer, MIT allows researchers to use the open source approach, and even manages their licenses, but this position is not part of an official policy.⁹⁰ The University of California system appears to have a highly complex set of positions on open source licensing. While some campuses, such as Berkeley and San Diego, are familiar with the open source model, other campuses are less so.⁹¹ Complexity is compounded by the fact that various important campuses, including Berkeley, have been actively fine- tuning their policies.⁹² As of early 2002, officials from the Berkeley office were quoted as criticizing the decision made by Berkeley a decade earlier, in 1992, to release as open source the Unix operating system and TCP/IP networking protocol.⁹³ At about the same time, Berkeley computational biologist Steven Brenner encountered difficulties in releasing his lab's software under an open source license.⁹⁴

As for the five schools that rank substantially higher in software/pure software patenting than in other patenting—the University of Washington, Georgia Tech, Carnegie Mellon, the University of Rochester, and the University of Illinois—the first three of these schools have unique characteristics that may explain their high levels of pure software patenting. At the University of Washington, a separate technology licensing office, now called the Digital Ventures Office, is responsible for managing pure software (and digital products more generally).⁹⁵ By 2005, the office employed seven full-time professionals as well as two to four part-time students.⁹⁶ As for Georgia Tech, its level of computer science R&D funding relative to other funding is quite high—it ranks fifth in computer science funding and only thirty-second in other funding. Finally, Carnegie Mellon ranks second in computer science funding and eighty-seventh in other funding. Carnegie Mellon is also home to the Software Engineering Institute (“SEI”), a research consortium founded by software firms.⁹⁷ Patents from SEI are assigned to Carnegie Mellon, and the software firms in the consortium receive an automatic, royalty-free license.⁹⁸

IV. University Software Ownership: A Policy Analysis

A. General Considerations

Our results indicate that universities have become active patentees of software, including pure software. They have clearly availed themselves of the opportunities afforded by the Bayh-Dole Act and the Federal Circuit decisions in the 1990s. Moreover, at least for our sample—which, because of prosecution lag time, terminates with patents filed around 2000—behavior with respect to total software patents and pure software patents is strongly affected by nonsoftware patent propensity. Although there are anomalies in the overall pattern—largely explicable through policies at particular universities that either strongly favor open source or favor a particular focus on software⁹⁹—the overall correlation is clear.

From a private point of view, this correlation may make sense, especially if the reason for the effect is, as some of our data suggests, economies of scale at the technology transfer office.¹⁰⁰ To the extent such scale economies exist, they are likely to lower the private marginal cost of patent acquisition.¹⁰¹ From a social point of view, however, patenting based on scale economies is quite problematic. Ideally, we would want decisions about whether to patent publicly-funded academic research to be based not only on the private marginal costs of patent acquisition, but also on whether a patent is needed to facilitate commercialization in a specific case, which is likely to vary across inventions and fields.

Stated differently, a lack of differentiation between software and other research could be problematic. As the data on the varying importance of patents as incentives in different industries suggests,¹⁰² and as case study research on the disparate development trajectories of university technologies specifically emphasizes,¹⁰³ the optimal mode of university-industry technology transfer is likely to vary by industry and invention.¹⁰⁴ More specifically, as indicated earlier, the theory espoused in the legislative history of Bayh-Dole—that patents and exclusive licenses are necessary to create incentives for firms to develop and commercialize “embryonic” university inventions¹⁰⁵—does not apply neatly in the software context where development costs are often low relative to other types of inventions.¹⁰⁶

Another major argument often advanced in favor of patents is that the prospect of licensing royalties induces university researchers to work with industry licensees and thereby transfer tacit knowledge necessary for commercialization.¹⁰⁷ Although this argument could in theory be compatible with exclusive or nonexclusive licenses, the assumption tends to be that an academic researcher would have sufficient time for only one exclusive licensee. However, in comparison to the life sciences, software (particularly pure software) is an area of invention where knowledge is likely to be relatively codified. Object-oriented programming is based on principles of modular design, and one of the reasons that open source methods of software production have been successful is that the development task can be broken up into modular pieces that are then reassembled.¹⁰⁸ So, the need for transfer of tacit knowledge may not be as pervasive as it is in the life sciences.

Indeed, in some well known cases, it appears that the university patent allowed the university and/or its exclusive licensee to extract rents from other firms without aiding in technology transfer. For example, in the prominent case of Eolas Technologies, Inc. v. MicrosoftCorp.,¹⁰⁹ it does not appear that Microsoft's commercialization was aided by the activities of the University of California/Eolas.¹¹⁰ Even worse, in Eolas, the university patent was arguably being used to “holdup” the commercializer—that is, to use the threat of injunctive relief to extract rents significantly larger than the inventive contribution made by the patentee (and thus also significantly larger than any licensing fee the university inventor could have charged ex ante).¹¹¹

B. The Problem of Holdup Litigation

In order to study the litigation question more systematically, we conducted two somewhat related empirical inquiries.¹¹² First, we used four different search methods to collect case studies involving university software litigation.¹¹³ Second, by running the patents in our database against the Derwent LitAlert database of patent cases filed, we determined whether the software patents in our sample were litigated at rates higher (or lower) than the nonsoftware patents in our sample.¹¹⁴

With respect to the latter inquiry, there was no statistically significant difference in rates of litigation between the software and nonsoftware patents in our sample.¹¹⁵ However, we did generate a number of case studies in which university software patents appear to have been used in a manner that hindered, rather than promoted, commercialization. Research Corporation Technologies (“RCT”), a firm that is the assignee of patents from the University of Rochester, has actively pursued litigation on a group of six patents covering the so-called Blue Noise Mask printing technology.¹¹⁶ This technology, which was developed at the University of Rochester and assigned to RCT under a technology evaluation and commercialization agreement dating back to 1953, allows for high-quality “half-tone” printing.¹¹⁷ It is used by firms ranging from Hewlett-Packard to Microsoft.¹¹⁸ The firm's web site emphasizes the patent suits it has brought against Hewlett-Packard, Epson, and Microsoft and “invites current and potential users of this landmark technology to contact [RCT] about licenses under RCT's patent rights.”¹¹⁹ The lawsuits against Hewlett-Packard and Epson were settled in 1999 and 2001, respectively.¹²⁰ The litigation against Microsoft is ongoing.¹²¹

MIT and its exclusive licensee, Akamai, were recently involved in lawsuits against two firms, Speedera Networks Inc. and C&W Wireless Internet Services, that allegedly infringed the MIT/Akamai patent on software for decreasing congestion and delay in accessing web pages on the Internet.¹²² The Akamai technology, which was launched in 1999 at the height of the “dot-com” boom, was similar to technologies developed by other firms before Akamai.¹²³ The district court granted a permanent injunction upon finding that the patent was valid and had been infringed.¹²⁴ On appeal, however, the Court of Appeals for the Federal Circuit determined that the broadest claims of the patent were invalid because the C&W software actually predated them.¹²⁵

In other cases, university patents have been asserted for large damage awards (usually for the statutory maximum of six years of past infringement) immediately before they are about to expire. For example, MIT and an exclusive licensee, Electronics for Imaging, have sued ninety-two firms, including Microsoft and IBM, alleging infringement of a patent covering image editing software.¹²⁶ This lawsuit was filed in December 2001, six months before the patent was set to expire.¹²⁷ The technology in question is a color imaging method that can be applied to any system that produces color pictures.¹²⁸ Similarly, in March and May of 2005, a few months before the relevant patent was due to expire, the University of Texas filed three lawsuits against a total of forty-two electronics manufacturers alleging infringement on a software patent that allows text messaging through a standard telephone keypad.¹²⁹ In some cases, the patent in question may not be due to expire immediately, but it is nonetheless relatively old. In 2002, for example, Cornell sued Hewlett-Packard over a 1989 patent obtained by a Cornell professor for a technique that accelerates a computer's processing speed.¹³⁰ Similarly, a 2001 suit by MIT against Lockheed-Martin involved a 1989 patent on systems for analyzing acoustic waveforms.¹³¹

In all of these cases, commercialization by firms other than the university licensee was going forward, and patent rights/exclusive licenses do not appear to have been necessary to facilitate “technology transfer.”¹³² Moreover, there is no evidence in these cases that the other firms' development efforts were “free-riding” on licensees' investments.¹³³ In fact, in at least three of these cases, the patent had not (apparently) been licensed at all.¹³⁴ Contrary to the spirit of Bayh-Dole, these patents simply allowed universities to extract rents and perhaps even holdup development efforts.¹³⁵

Attempts to extract rents from firms that have commercialized successfully may be a particular concern where the case is ultimately very weak. Thus, it is notable that in a number of litigated cases, the university's argument has been unequivocally rejected.¹³⁶ For example, the University of Texas recently lost a case involving patents on positron emission technology (“PET”), a medical imaging technology that uses gamma rays to detect cancer, heart disease, brain disorders, and other health conditions.¹³⁷ The University of Texas claimed that CTI Molecular Imaging Inc., a leading provider of PET equipment, infringed two of its patents.¹³⁸ In that case, both the district court and the Federal Circuit held that the defendant did not infringe.¹³⁹ Similarly, in a recent suit brought by the University of California and its exclusive licensee over patented software for eliminating edge artifacts when compressing digital images, both the district court and the Federal Circuit found that the patent in question was neither invalid nor infringed.¹⁴⁰ In 2003, a lawsuit by the University of Illinois against Fujitsu on software patents relating to plasma display panels resulted in a summary judgment determination that the relevant patent claims were invalid.¹⁴¹ In contrast, our search found only one case in which a university's assertions regarding its patented software were largely vindicated by the court system, either in a final district court decision that was not appealed or in an appellate court decision.¹⁴²

Even if patenting and exclusive licensing of software do not facilitate commercialization per se, one might argue that exclusive licenses to university patents are useful for software start-ups, and promoting such start-ups is a socially valuable goal.¹⁴³ But, even assuming that the “generating start-ups” argument has merit, the force with which it applies to software is not clear. While most small biotechnology firms that receive venture financing have patents, the available empirical evidence indicates that most software start-ups that receive venture financing, particularly in the first round, do not have patents.¹⁴⁴ Moreover, though there is some evidence that small software firms with patents tend to fare better than firms without such patents,¹⁴⁵ the evidence also indicates that the disparity is much less significant than the disparity between biotechnology firms with and without patents.¹⁴⁶ At a minimum, the “generating start-ups” argument for patenting software is less compelling than it is for biotechnology. In any event, in many of the cases we found, the university was suing on its own behalf; it did not appear to have found a licensee.¹⁴⁷

Of course, litigated cases are not likely to be representative of all university software patents. There are certainly cases of successful commercialization where the firm in question did have an exclusive license to a university patent. The company Google, which was the exclusive licensee of a patent assigned to Stanford,¹⁴⁸ is a prominent example, as is RSA Security, the exclusive licensee of various data encryption patents assigned to MIT.¹⁴⁹ Moreover, as noted earlier, university software patents do not appear to be litigated at rates significantly different from those of university patents in other fields.¹⁵⁰ Thus, we cannot state unequivocally that the incidence of holdup in the software patent context is higher than that for university patents in other fields.

Nonetheless, there is reason to believe that holdup is more likely in software than other fields, mainly because patents and exclusive licenses are less likely to be important for commercialization in this field than others. To put the point another way, the ratio of false positives (patenting and giving an exclusive license when it is not necessary for commercialization) to false negatives (failing to patent and give an exclusive license when it is necessary for commercialization) is likely to be higher in software than in the life sciences. In this regard, it bears mention that some of the most successful cases of software commercialization have not involved patents and exclusive licenses. For example, a number of unpatented Stanford programs have been widely adopted by the industry: both MINOS, a linear and nonlinear optimization program, and Genscan, a gene structure prediction program, have been used (via a copyright license) by dozens of different commercial firms.¹⁵¹

V. The Way Forward

As the Stanford example suggests, models of successful software commercialization absent patents are possible. More generally, the future of university software ownership may not be one in which universities are using patents as holdup opportunities.

A. The Impact of eBay v. MercExchange and In re Bilski

As an initial matter, the holdup threat that patentees can wield may have been mitigated by the Supreme Court's decision in eBay Inc. v. MercExchange, L.L.C.¹⁵² In that case, the Court made it clear that permanent injunctive relief is not automatic upon a finding of patent infringement.¹⁵³ Rather, the plaintiff must prove each element of the traditional four-factor test for permanent injunctive relief: (1) that it has suffered irreparable harm; (2) that remedies available at law, such as monetary damages, are inadequate to compensate for that harm; (3) that the balance of hardship between the patentee and infringer argues in favor of a remedy in equity; and (4) that the public interest would not be disserved by a permanent injunction.¹⁵⁴ Although the Supreme Court opinion explicitly addresses only permanent injunctive relief,¹⁵⁵ it has been interpreted by the lower courts as raising the burden faced by plaintiffs in requests for preliminary injunctive relief.¹⁵⁶ The application to preliminary injunctive relief is important because, as was indicated by the famous $612.5 million settlement that RIM, the maker of the BlackBerry device, paid on patents that had been seriously questioned by the PTO illustrates,¹⁵⁷ even the threat of a preliminary injunction—before validity and infringement have been proven—can be used to extract large settlements.

Lower court cases interpreting eBay, particularly in the context of permanent injunctive relief, have tended to emphasize that irreparable harm should be presumed, and injunctive relief is thereby available, where the patentee and the infringer are direct competitors; in that case, infringement is likely to cause the patentee to lose market share and profits.¹⁵⁸ In an instructive opinion, the eBay district court noted on remand that irreparable harm may also be found where a patentee is seeking to develop its patent in partnership with others (the classic Bayh-Dole rationale).¹⁵⁹ In contrast, where the patentee secures revenues by approaching firms that have already developed, so as “to maximize the value of a license,” monetary damages should be sufficient to compensate for infringement.¹⁶⁰ Thus, at least in those situations where the university patent is being asserted without a partner licensee that practices the patent in competition with infringers, one might suppose that the university's post-eBay bargaining leverage would be reduced.

The question of how universities should be treated for purposes of injunctive relief was the subject of a recent Federal Circuit appeal, Commonwealth Scientific and Industrial Research Organisation v. Buffalo Technology Inc.¹⁶¹ Although the patentee in that case, Commonwealth Scientific and Industrial Research Organisation (“CSIRO”), is an Australian research institute, not a U.S. university, it generates and licenses technology in precisely the same manner as do U.S. universities.¹⁶² Moreover, the patent in question was a software patent that covers wireless LAN technology, specifically a number of 802.11 standards.¹⁶³ In CSIRO I and related cases, CSIRO asserted its patent against not only Buffalo Technology, but also Toshiba, SMC, 3Com, and Microsoft.¹⁶⁴ The district court decision granting injunctive relief to CSIRO emphasized the eBay court's reluctance to categorically rule out injunctive relief for universities.¹⁶⁵ The court then went further, ordering injunctive relief on the theory that a failure to grant such relief would create irreparable harm by making it difficult for CSIRO to license its patents and thus support its research enterprise.¹⁶⁶ Therefore, the extent to which university patentees could continue to exercise the threat of holdup would appear to have turned on the Federal Circuit's resolution of CSIRO I. However, the Federal Circuit remanded the case on invalidity grounds without ruling on the issue of injunctive relief.

B. Revenue Generation Without Holdup

Contrary to the district court's theory in CSIRO II, university interest in revenue generation can be achieved without the innovation-chilling threat of holdup. Specifically, the alternative of ex ante nonexclusive licensing at reasonable rates (which CSIRO allegedly did not attempt to do despite a promise to the Institute of Electrical and Electronics Engineers (“IEEE”) to license on “reasonable and nondiscriminatory” terms¹⁶⁸) could be a mechanism for satisfying university interests while minimizing potential harms.¹⁶⁹ Although nonexclusive licenses are something of a tax on commercialization, a small royalty rate is unlikely to deter most commercialization.

Sound practical reasons may, however, counsel against nonexclusive licensing of patents. Although, as noted earlier, economies of scale may reduce prosecution costs for large technology transfer offices,¹⁷⁰ the prosecution of patents still requires some investment of money and staff time. While exclusive licensees typically pay patent application costs, nonexclusive licensees generally do not.¹⁷¹ Notably, however, software (particularly pure software) differs from virtually every other university invention in that patents do not have to serve as the foundation of a nonexclusive licensing scheme. Copyright, which attaches without cost upon creation of the software, will do the job.¹⁷² For this reason, at least some technology transfer offices state that they are beginning to refrain from seeking patents.¹⁷³ At the University of Washington, for example, the Digital Ventures office says that it uses patenting only if there is a real need to improve the technology (presumably via an exclusive license).¹⁷⁴ Digital Ventures has also convinced start-ups to “go without a patent.”¹⁷⁵ For example, MINOS, which is available via a nonexclusive copyright license, is one of Stanford's largest money generators.¹⁷⁶

Because copyright has been interpreted by the courts to cover little more than source code,¹⁷⁷ it generally does not, by itself, confer a great deal of power. For this reason, commentators concerned about the negative effects of strong proprietary rights have typically focused on patents. Even so, firms appear to be willing to license software because they do not want to bear the costs of independent creation.

On the other hand, even nonexclusive copyright licensing with relatively low fees can be problematic for nonprofit researchers. Thus, universities that want to balance the goal of academic access with revenue generation, such as the University of Washington, are assessing which licenses and royalty structures are appropriate for which situations.¹⁷⁸ The technology transfer offices at the University of Washington, the University of California at Berkeley, and Stanford University have all embraced the idea of dual licenses that give relatively inexpensive access to the nonprofit sector but allow for revenue generation from the commercial sector.¹⁷⁹ Indeed, Stanford's lucrative MINOS business software program is available via a dual copyright license—the commercial use costs more than ten times as much as the academic use.¹⁸⁰

C. Evolving Policies at “Pro-Patent” Institutions?

Our research has also generated some evidence of evolving policies at institutions that have historically sought large numbers of software patents. In 2004, the University of California at Berkeley announced a policy that appeared to give a relatively strong endorsement to the open source model. In this new policy, the Berkeley Technology Transfer Office states that it will work with researchers who want to release their code under an open source model.¹⁸¹ Although it is not clear whether researchers have the final say in cases of conflict, the new policy appears encouraging towards open source. Berkeley has also developed, and encourages researchers to use, an “academic” license, under which source code is made available free of charge to academic institutions and nonprofits, while being made available for a fee to commercial users.¹⁸² In general, the new Berkeley policy appears to encourage the possibility of releasing software under different types of licenses for different purposes.

Moreover, although the University of Washington, Georgia Tech, and Carnegie Mellon have a significant software patent presence, this level of patenting may not represent a continuing pattern. Current technology transfer officials assert that they promote other models of software ownership. In fiscal year 2005, the University of Washington's Digital Ventures office generated ninety percent of its revenue from copyright licensing.¹⁸³ The Digital Ventures web site features a large variety of unpatented software that can be secured for commercial use through an online license with a standard rate.¹⁸⁴ Like Berkeley, the University of Washington also encourages “forking” in its licenses—licenses in which software (and underlying source code) is made available free of charge for certain uses and available for a fee for other uses.¹⁸⁵ Officials at Georgia Tech's technology licensing office note that they oversee dozens of open source-type software licenses each year and that, given the short technology life cycle in the software industry, algorithm patents are often not very valuable.¹⁸⁶ Finally, at Carnegie Mellon, the current technology transfer team states that it almost never patents unless another entity is willing to pay for such patenting.¹⁸⁷

Furthermore, information technology can substantially facilitate the task of low-transaction cost, nonexclusive licensing. Universities such as Stanford already make software like MINOS available through simple web-based interfaces.¹⁸⁸ Several dozen universities are currently participating in the Kauffman Foundation's web-based I-Bridge project, which aims to reduce the transaction costs associated with licensing: many of the projects available on the I-Bridgesite represent software.¹⁸⁹ Additionally, some large technology firms have decided to sponsor software projects at universities under an explicit “open collaboration” model.¹⁹⁰ In 2005, four technology firms (IBM, Hewlett-Packard, Intel, and Cisco) announced a set of “Open Collaborative Research” principles under which certain types of sponsored research in software would be made freely available.¹⁹¹ IBM is now embarking on specific software projects at seven universities under the rubric of these principles.¹⁹²

D. The Role of Open Source

The related question of using copyright to promote open source within the university is an interesting one. As we have noted, some universities have long embraced open source, and these universities tend to have smaller numbers of pure software patents.¹⁹³ But even among technology transfer officials sympathetic to the goals of open source, some mention difficulties that open source may create in the university setting.¹⁹⁴ For example, faculty may prefer open source as a method of distribution, not because of ideological commitment, but because open source-related consulting revenues, unlike licensing royalties, do not have to be shared with the university.¹⁹⁵ Indeed, at least one prominent technology transfer officer (who preferred to remain anonymous) believes that some faculty make software open source for the purpose of attracting widespread interest, but have every intention of asserting proprietary rights over the source code at some later point.¹⁹⁶ Additionally, software is often developed by groups, and technology transfer officers sometimes find themselves in the middle of disputes among group members about the best open source mechanism to use (or, indeed, whether open source should be used at all).¹⁹⁷ Technology transfer officers are also wary that particular types of open source licenses will conflict with obligations to sponsors, including the federal government under Bayh-Dole.¹⁹⁸ Thus, to the extent funding agencies are interested in an open source approach because they think such an approach is likely to produce better software, they need to be aware of possible institutional impediments.

One final point bears emphasis. Because our quantitative data concludes with patents filed in the late 1990s, we cannot say that university software holdings have continued to grow. Indeed, as noted, technology transfer officers from universities with significant software patent holdings, such as the University of Washington and Carnegie Mellon, state that they are now quite receptive to nonpatent-based modes of technology transfer.¹⁹⁹ Whether university patenting and licensing practices are currently different from practices in the late 1990s would be a valuable subject for further research. Even if filing practices have changed, however, the existing stock of software patents may represent a lucrative source of revenues when asserted against firms that have commercialized successfully. The extent to which such holdup strategies are successful may depend on how the CSIRO line of cases are resolved.

Conclusion

In this Article, we have undertaken the first systematic investigation of university software ownership and litigation. We found that software patents represent a significant, and growing, percentage of university patent holdings. Moreover, our qualitative results and research from a complementary econometric analysis, suggest that universities have followed a “one size fits all” approach: those with higher patent propensities in other fields also tend to patent more software. From a policy standpoint, this finding suggests a problem. Software—and particularly pure software—is likely to follow a different commercialization path than invention in other fields. Therefore, patenting and exclusive licensing of software may yield a higher proportion of situations where the exclusive licensee uses a patent to hold up an entity that has successfully commercialized without the need for an exclusive license. Moreover, even if the goal is not promoting commercialization per se, but promoting start-ups, exclusive patent licenses are not necessarily critical to that goal.

To be sure, university practices with respect to software are not uniform. For example, universities that have policies friendly to open source are less likely to patent pure software. More generally, a fair number of universities—including universities with large numbers of software patents—may now be moving away from patenting and exclusive licensing of software.²⁰⁰ However, the lingering impact of a substantial university patent portfolio, some of which has been licensed exclusively, may continue to be felt in socially unproductive litigation.

Table 9. Departmental Origin of Inventors on Software Patents Issued to the Top Fifteen Recipients of Software Patents in Issue Year 2002.

Department	N	Percent	Cumulative Percent
EE/CS	47	9.67	9.67
Electrical Engineering	47	9.67	19.34
Computer Science	42	8.64	27.98
Neuroscience	23	4.73	32.72
Robotics Institute	23	4.73	37.45
Medical Physics	20	4.12	41.56
Radiology	19	3.91	45.47
Lawrence Livermore	15	3.09	48.56
Chemistry	13	2.67	51.23
Mechanical Engineering	12	2.47	53.70

Counts and percentages calculated at the patent-inventor level

Appendix.

Questions for Technology Transfer Officers

1. Is there an official university policy regarding ownership and dissemination of software and software-based inventions? If yes, what is it?

2. Are there more informal guidelines or norms that govern ownership and dissemination of software or software-based inventions? If yes, what are they?

3. How does the technology transfer office decide whether it should seek a patent on software or software-based inventions?

4. What is the university's policy, if any, towards open source software?

5. Are there any trends or changes in university policy towards ownership of software or software-based inventions that are worthy of note?

6. Have you had any conflicts with faculty members regarding how software or software-based inventions should be owned or disseminated? In cases of conflict, who makes the determination?

Questions for Academics/Graduate Students

1. Are you aware of an official university policy regarding ownership and dissemination of software and software-based inventions? If yes, what is it?

2. Are you aware of informal guidelines or norms that govern ownership and dissemination of software or software-based inventions? If yes, what are they?

3. Do you know how the technology transfer office decides whether it should seek a patent on software or softwarebased inventions?

4. Are you aware of any university policy towards open source software? If yes, what is it?

5. Are there any trends or changes in university policy towards ownership of software that are worthy of note?

6. Have you had any conflicts with the technology transfer office regarding how software or software-based inventions should be owned or disseminated? In cases of conflict, who makes the determination?

7. Do you produce or use open source software?

8. Do you generally disclose software or software-based inventions to the technology transfer office?

Table A1. Comparison of Bessen-Hunt Keyword-Based Classification of Software Patents with Our Classification.

	Our Classification
Bessen-Hunt Classification	Non SW	SW	Total
Non SW	2,325	202	2,527
	92.01	7.99	100
	91.32	51.01	85.89
SW	221	194	415
	53.25	46.75	100
	8.68	48.99	14.11
Total	2,546	396	2,942
	86.54	13.46	100
	100	100	100

Table A2. Comparison of Graham-Mowery IPC-Based Classification of Software Patents with Our Classification.

	Our Classification
Graham-Mowery IPC Classification	Non SW	SW	Total
Non SW	2,544	341	2,885
	88.18	11.82	100
	99.92	86.11	98.06
SW	2	55	57
	3.51	96.49	100
	0.08	13.89	1.94
Total	2,546	396	2,942
	86.54	13.46	100
	100	100	100

Table A3. Comparison of Graham-Mowery USPC-Based Classification of Software Patents with Our Classification.

	Our Classification
Graham-Mowery USPC Classification	Non SW	SW	Total
Non SW	2,541	328	2,869
	88.57	11.43	100
	99.80	82.83	97.52
SW	5	68	73
	6.85	93.15	100
	0.20	17.17	2.48
Total	2,546	396	2,942
	86.54	13.46	100
	100	100	100