Today many are turning to biometrics technology - the measurement and analysis of biological data such as fingerprints, eye retinas and voice patterns for the specific purpose of identification. But while interest in biometrics for all sorts of information security and physical access applications is on the rise, mass market adoption of the technology has not yet occurred, primarily due to cost and reliability issues.
While there are many different possible biometric approaches, let's focus on fingerprint authentication, which is perhaps the closest to attaining widespread use. To date, there have been several factors that made the cost of implementing fingerprint authentication prohibitive. Beyond the cost of the device itself, which is now dropping rapidly, there are also integration costs, made more complicated by the lack of a unifying standard, and costs related to maintaining a secure database for the fingerprint data.
Problems with reliability have also created a barrier to widespread adoption. The reliability of various fingerprint sensors has typically been hampered by environmental factors such as contamination or moisture, and many sensors are susceptible to damage caused by electrostatic discharge when the finger comes into contact with the sensor. These challenges have driven the rapid evolution of a new generation of sensors using different materials and circuitry that circumvent many of the problems that plague earlier devices. To more fully understand where these developments are leading, it's helpful to have a little background information about how fingerprint authentication actually works.
What is a Fingerprint?
Fingerprint authentication is based, obviously enough, on an individual's fingerprint, but the print is used in an entirely different manner than in forensic fingerprinting. To appreciate the distinction and understand modern fingerprint authentication, we need to look at the anatomy of a fingerprint.
Fingerprints are composed of ridges, the elevated lines of flesh that make up the various patterns of the print, separated by valleys. Ridges form many patterns, including loops, whorls, arches and more (Figure 1). Minutiae are discontinuities in the ridges, which can take the form of various ridge endings, bifurcations (forks), crossovers (intersections), and many others.
Figure 1. Typical fingerprint features.
Fingerprint authentication is based on a subset of features selected from the overall fingerprint. Data from the overall fingerprint is reduced, using an algorithm application usually unique to each vendor, to extract a dataset based on spatial relationships. For example, the data might be processed to select a certain type of minutiae or a particular series of ridges. The result is a data file that only contains the subset of data points - the full fingerprint is not stored, and cannot be reproduced from the data file. This is in contrast with the forensic fingerprinting that most people associate with the term 'fingerprinting,' which is based on the entire fingerprint.
A typical fingerprint authentication data file is less than 1,000 bytes, and is used for one-to-one or one-to-few verification, which takes a few seconds. Modern forensic fingerprinting, on the other hand, creates files on the order of 250Kb for each complete fingerprint; these files are used in large-scale, one-to-many searches with huge databases - a process that commonly requires hours for verification.
Fingerprint Authentication in Operation
In use, fingerprint authentication is very simple. First, a user enrolls in the system by providing a fingerprint sample. The sensor captures the fingerprint image, which is then interpreted by an algorithm and the representative features extracted to a data file. This process can take place either on a host computer or a local processor (in applications such as cellular handsets). The data file then serves as the user's individual identification template. During the verification process, the sequence is repeated, generating an extracted feature data file. A pattern-matching algorithm compares the extracted feature data file to the identification template for that user, and the match is either verified or denied. State-of-the-art processor, algorithm and sensor systems can perform these steps in a second or two.
Modern Fingerprint Authentication Technology
Fingerprint authentication can be based on optical, ultrasound or silicon sensors. Optical technology is the oldest and most widely used, and is a demonstrated and proven technology, but has some important limitations. Optical sensors are bulky and costly, and can be subject to error due to contamination and environmental effects. Ultrasound, utilizing acoustic waves, is still in its infancy and has not yet been widely applied. We are focusing here on silicon sensor technology, introduced in the late 1990s, which is being increasingly applied, and offers some important advantages.
Silicon sensors are based on a two-dimensional array of cells, as shown in Figure 2. The size and spacing of the cell is designed such that each cell is a small fraction of the size of the ridge spacing. Cell size and spacing are generally about 50 microns, yielding a resolution of up to 500dpi, the FBI's image standard. When a finger is placed on the sensor, the image is captured by activating transistors that underlie each individual cell. Each cell individually records a measurement from the point on the finger directly above the cell.
Figure 2. Typical silicon sensor array.
The most commonly recorded measurement is the distance, or spacing, between the sensor surface and that part of the finger directly above it (Figure 3). However, the measurement can also be based on pressure rather than distance, which has some inherent advantages. The two methods are described below.
Figure 3. Output from a distance-measurement sensor.
The set of distance measurements from all cells is integrated to form a raw, gray-scale fingerprint image as shown in Figure 4. Fingerprint imaging using a continuum of distance measurements results in an 8-bit gray-scale image, with each bit corresponding to a specific cell in the two-dimensional array of sensors. The extreme black and white sections of the image correspond to low and high points on the fingerprint. Only the ridge points on the fingerprint are of interest, since they correspond to the ridges on the fingerprint that are for identification. Therefore another algorithm must be used to deconvolute the 8-bit gray-scale image into a 1-bit binary image.
Figure 4. Gray-scale fingerprint image based on distance measurements.
A feature extraction algorithm is then applied to the fingerprint to extract the specific features that make up the individual's unique data file. This data file serves as the user's individual identification template, which is stored on the appropriate device.
A significant source of error in distance measurement can be introduced by the presence of dirt or grease, which can be falsely interpreted as high points, resulting in errors in minutia extraction and the subsequent feature matching process.
The principle of pressure sensing is that when a finger is placed over the sensor, only the ridges come in contact with the individual sensor cells, and no other part of the finger contacts the sensors. As a result, only those sensors that experience the pressure from the ridges undergo a property change, such as a change in resistance. Pressure sensors are architecturally similar to other silicon sensors in terms of cell size and spacing, and therefore offer similar resolution.
One of the principal differences between pressure and distance sensing methods is that pressure sensors directly generate a 1-bit binary image. While there is more information in an 8-bit gray-scale image, much of it is extraneous and must be filtered out. Although the resistance value generated by a pressure sensor is an analog value, the difference between the resistance in the pressed and unpressed states is large enough that, with an appropriate threshold setting, one can easily distinguish between the presence or absence of a ridge with high resolution and accuracy. This is in effect a 'digital' response - the sensor either records a ridge or a valley (Figure 5). In contrast, distance measurement techniques generate a continuum of measurements or a gray-scale, which must be corrected for noise reduction, gray-scale adjustment, gain and sensitivity adjustment. The result is that pressure measurement can offer improved accuracy of ridge and valley detection.
Figure 5. Output from a pressure-measurement sensor
Pressure sensors can also be considerably less sensitive to interference from dirt and grease on the finger or the sensor, wet or dry fingers, and other effects. The moisture, grease and grime that might be on a finger are normally present as thin layers on the surface of the skin, and there is usually no effect on a pressure-based sensor. With a distance-based measurement, these thin layers can cause significant distortion in the resulting image output. An example of a fingerprint image under wet and dry conditions for a pressure sensor versus a distance sensor is shown in Figure 6.
Figure 6. A fingerprint image under wet and dry conditions for a pressure sensor versus a distance sensor.
Manufacturing and Cost
Distance measuring sensors employ active switching devices for addressing each cell, such as diodes or transistors that are built into each cell. These are normally built using complementary metal-oxide semiconductor (CMOS) technology, which dictates that the entire sensor be built on a silicon substrate. However, considering that the fingerprint array is normally about 16mm by 18mm, only about 20 to 25 of these devices can be made on a 6-inch silicon wafer. Since the cost of processing a silicon wafer is typically in the range of $500 to $600, the cost of a device is unlikely to fall much below $20. This clearly limits the applicability of this type of sensor for many markets such as computing, PDAs, cell phones and smartcard readers.
Pressure sensors, on the other hand, can utilize a passive cell-addressing method, which is integrated with the cell array, but is not built into each cell. This allows pressure sensors to be manufactured using thin film technology, which offers several significant advantages compared to silicon wafer semiconductor methods. Thin film technology is substantially cheaper than conventional semiconductor manufacturing methods because devices can be produced on large panel-like substrates rather than 6 inch or 8 inch diameter single crystal silicon wafers. When 1,000 devices can be manufactured on a single panel (as opposed to 25 devices on a silicon wafer), the fully assembled manufacturing cost per device drops to under $5.
Thin film methods are also not restricted to silicon substrates. Sensors can be produced on glass, ceramics, plastics and other materials. These alternative substrates reduce cost and allow greater flexibility for integration with all types of devices. Furthermore, sensors manufactured by thin film methods are much less sensitive to electrostatic discharge (ESD). ESD damage is common on CMOS-based circuits unless they are protectively packaged and sealed when they are incorporated into the products that use them. A fingerprint sensor cannot be protected in this manner since it requires direct contact by the user, and is open to the environment. A CMOS device is therefore exposed to the risk of ESD damage with every use. Elimination of the CMOS-based active circuitry drastically reduces this susceptibility and increases reliability.
The overall market for biometrics security has been forecast to grow significantly, perhaps reaching over $1 billion within the next few years, and fingerprint authentication is widely seen as the most reliable and convenient method. As a result, efforts have been increasing to develop, integrate and bring to market reliable and cost effective fingerprint authentication solutions. Modern silicon sensors are rapidly solving the cost and reliability problems that have limited the widespread adoption of fingerprint authentication as a biometric security application. New sensors that are based on thin film-manufacturing methods and which employ novel measurement techniques have a promising future in the face of a growing demand for simple and inexpensive security tools.
S. K. Ganapathi is president and CEO, Fidelica Microsystems (www.fidelica.com), which he founded in October 1999.