A Direct Fingerprint Reader
|Authors:||Dale Setlak, Senior Principal Engineer, Harris Semiconductor|
|Karl McCalley, Vice President of Systems, AuthenTec, Harris Corporation|
|Steve Wilson, President, Clearpoint Corporation|
|John Schmitt, President, Schmitt Associates Inc.|
Presented at the 1996 CardTech/SecureTech Show
The ability of computers to directly and reliably identify individual human beings is a key enabling technology to a whole new class of electronic systems, devices, and applications. A direct-sensing electronic fingerprint reader, currently under development, may satisfy the need for a small, inexpensive, reliable, and easy-to-use sensor for personal electronic identification.
Human fingerprints have been used as a form of identification and authentication for thousands of years, dating back to when the Chinese certified the legality of documents by making a finger impression in sealing wax. Today fingerprints are well established as unique and persistent identifiers of individuals. These characteristics make fingerprints potentially useful as a means for electronic systems to reliably identify human beings.
Over the last couple of decades, advances in computers, software, and other technologies have made it possible to store fingerprint data in electronic databases and automatically search for matches. Although there is some use of fingerprints in other fields, such as government ID systems, law enforcement remains the most prominent application of fingerprint technology. Wider use of use of fingerprints as a form of personal identification has been severely limited by several factors:
The oldest and still the most common way to capture a fingerprint for computer processing is electronic scanning of a card with an ink-and-roll image of a print. This capture process is slow, messy, and prone to error. Smudging and smearing are common problems that go unnoticed at the time of ink capture. Feedback from a computer about the acceptability of the inked image is not available until a later time when the card is electronically scanned.
Optical live scan units have difficulty reliably capturing a good quality image. One common technique used, Frustrated Total Internal Reflection (FTIR), requires close contact between the finger and the glass platen for a good image. Dry skin, stains, and dirt all too often interfere with the image.
Currently available live scan devices are too large to readily integrate into computer terminals and appliances. Typically about the size of a desk telephone, live scan units are as large as the devices in which they might be useful, such as credit card authorization terminals or laptop computers. The size is determined by the required spacing of the optical components in the optical systems or the size of the motors and mechanical assemblies of other types of scanning systems. Power supplies also add size and weight, sometimes nearly as much as the scanner itself.
Human attendants to perform the ink-and-roll process are expensive. Optical live scan devices typically cost well over a thousand dollars, especially when the cost of a video frame grabber or other computer interface is considered. The newest generation of live scan devices using ultrasonic imaging cost as much as a personal computer, placing them well above the price point where they will see widespread commercial application. The total system cost is actually even higher when the cost of the image processing computer is included.
Beyond the obvious speed problems with ink-and-roll processing, even electronic live scan devices are slow to emit data useful to a computer application. Some models take several seconds just to capture image data and transmit it to the computer for subsequent image processing and minutia extraction. The processing of the image takes seconds before the minutia are generated and presented to the application. Specialized quick-responding recognition algorithms lack the ability to distinguish individuals amongst sufficiently large databases.
There is growing interest in electronic personal identification and authentication technology. Networked systems, electronic commerce over the Internet, private E-mail over the Internet - all of these have personal authentication and identification needs. These are examples of a growing class of computer / network applications in which the system must certify the identity of its user before it proceeds. These applications can vary widely in size and scope, and run the gamut from life-critical to entertaining -- from computerized patient medical record systems, to automated border control, to on-line computer dating services. The common denominator is the need for a simple, convenient, reliable, and inexpensive way for the system to identify a person, without the assistance of a second person.
Though there are many ways to verify a human identity to some degree, biometric identification is considered highly desirable. Biometrics offer a strength of authentication not possible with passwords, cards or tokens -- all of which can be stolen and used by unauthorized parties.
There are many characteristics of human beings that can be used for biometric identification. These include:
The fingerprint biometric stands out among these possibilities. Fingerprints are well understood and widely accepted as a unique form of human identification. Fingerprints are known to be naturally stable throughout a lifetime and fingers are readily accessible for non-intrusive sensing. Time-tested fingerprint comparison methods are known to be capable of distinguishing individuals reliably within extremely large populations.
The wider use of fingerprints in electronic systems, however, will require more effective and economical ways of capturing fingerprints "live."
There are many physical phenomena that could be used to sense the fingerprint pattern. These range from simple mechanical contact devices to a wide range of energy-field and radiation sensing devices. There are several types of live scan devices available today using various physical means to read the subject's fingerprint. Examples include:
Optical Frustrated Total Internal Reflection
Ultrasonic Scanned Reflection
Thermal Under Development
Another physical effect that can be used to determine the shape of the ridges and valleys of a fingerprint is an electric field -- an e-field. Electric fields form in the regions surrounding any accumulation or movement of electric charge. Our bodies are constantly bathed in electric fields that are generated both external to the body, and by the body itself.
An electric field sensing system applies a small electric field close to the skin surface through a conductive element called an excitation antenna. The body serves as a ground reference for this field. The electric field voltages at various points between the excitation antenna and the skin reflect the shape and composition of the skin. An array of tiny sensing elements positioned between the skin and the excitation antenna sense the spatial variations in the field voltages. Figure 1. Illustrates these key concepts.
Viewed from a circuits perspective, the skin of the subject acts as one plate of a capacitor. An array of sensors between the skin and the excitation antenna sense variations in the field due to the varying distances to the ridges and valleys of a fingerprint. The elements of this kind of system can be fabricated at the necessary scale on a silicon chip using typical semiconductor manufacturing processes.
Capacitance-based fingerprint readers have been attempted before. The capacitance based sensors can be viewed as a simplified form of an electric-field sensor. The devices share many of the same engineering challenges:
The capacitance between neighboring pixel sensors can be greater than the effects due to the skin, reducing the signal to unusable levels and effectively blinding the sensing elements -- like the glare from an automobile windshield.
Capacitive sensors located over a fingerprint valley have a tendency to sense the neighboring ridges rather than the valley between. This defocusing effect reduces the ability of the sensors to resolve features -- effectively blurring their field of view.
The pixel sensing elements are small, necessarily small enough to detect fingerprint ridges and valleys. The absolute levels of the signals is therefore tiny and easily swamped by noise or spurious signals from any source. We were most concerned about two sources: 60Hz power line noise picked up by the user and electrical noise from within the sensor, important here because of the tiny signals to be detected. For example, the signal leads used in several earlier attempts at capacitive fingerprint sensing picked up noise and distortion, reducing the signals to unusable levels.
A human finger is rather hostile when viewed by a naked semiconductor chip. Electrostatic discharge, salt from sweat, and physical wear are potentially lethal to a semiconductor sensing chip.
Experience with existing live scan devices has shown a wide variety of temporary skin conditions and sensing artifacts that can significantly affect the sensed image, and confuse the subsequent steps of ridge detection and minutia extraction.
The AuthenTec FingerLoc sensor is a breakthrough. A series of innovations work together to resolve the above issues.
Specially designed antenna structures and novel electronic circuitry for the electric field generating and sensing elements have been designed, that control and focus the sensing region of each sensing element. The unique design of these structures overcomes the effects of sensor fringing and parasitic capacitances.
The electric field sensor's physics allows it to develop clean ridge images from very dry, cracked, worn, or otherwise damaged fingers that are very difficult to image effectively using optical techniques.
Signal noise has been a key problem in previous attempts to construct capacitive fingerprint sensors. In the design under current consideration, the signals on the sensor elements are measured by low-noise data acquisition circuitry located on-chip. Signal leads are short and noise pickup and sensor loading are minimal. The signals are subsequently processed by high-performance noise reduction circuitry. The design enhances signal-to-noise ratios to levels well in excess of that required for good image feature extraction.
Harris put its long experience with signal processing to use in the design of an "intelligent" sensor array. The on-chip pixel electronics performs complex image enhancement processing in real time. The array electronics automatically filter out the majority of the temporary skin conditions and sensing artifacts that are typically seen in live scanned fingerprint images. The output is a clean high-contrast image, eliminating the need for the time consuming digital image processing that is prerequisite to most high resolution feature extraction algorithms when using traditional optical sensors.
Conventional semiconductor packaging is designed to prevent the chip from being touched. The e-field sensor packaging must permit the finger to contact the semiconductor chip without damaging it. Harris has developed packaging for this device that safely presents the sensing surface to external contact while protecting the circuitry from damage or contamination. The package also allows the device to be mounted securely, able to withstand the forces of many finger presses. The chip can be supported independently from associated circuit boards, to prevent transferring the finger pressure into circuit board flexing.
The chip will be exposed to human sweat, various other contaminants, and various cleaning and disinfecting agents. Sodium ions are particularly destructive to silicon integrated circuits. They migrate through silicon, leaving a trail of damage through the semiconductor crystal lattice. A specially designed surface coating is under development to protect the chip from these kinds of chemicals.
The size of the component is less than one half of a cubic inch, several hundred to a thousand times smaller than other live scan devices. This small size means that FingerLoc sensor can be placed almost anywhere that a button or switch can be placed. It can be integrated into point-of-sale terminals, door locks, and TV set-top cable access units, to name a few examples. Power consumption is so small that the device can be used even in battery powered devices such as laptop computers or other appliances.
The image can be captured and processed in approximately 100 milliseconds. This speed is independent of image size due to the coherent nature of the sensing and the parallel nature of the integral image processing. After the on-chip processing, the image can be presented to the system as a block of memory, one byte per pixel representing an eight bit gray scale value. The processor performing the minutia extraction can operate directly on this memory array. In this case there is no data interface transfer time; and no serial, SCSI, or video frame grabber interface required. However, peripheral style interfaces (such as serial, SCSI, or PCI) can be provided where demanded by higher level integration products.
The cost of the FingerLoc sensor will be roughly an order of magnitude less than existing live scan products. An e-field sensor can be manufactured using semiconductor manufacturing processes. The low cost is a result of the large volume automated manufacturing techniques that are available for semiconductor components. Unlike other types of scanners, the electric field fingerprint sensor requires no critical alignment of components, no focusing of lenses, and has no moving parts.
Most semiconductor devices achieve cost reductions by becoming smaller, requiring less silicon wafer surface area to achieve the necessary functionality. A fingerprint reader's size is necessarily determined by the size of the human finger, therefore greater circuit density will not translate to reduced chip size and. Manufacturing process improvements, however, can reduce the cost per square inch and reduce the defect rate, improving yield which lowers the cost. Therefore the cost of the sensor will decline over time, though not as quickly as with other semiconductor components.
The e-field sensor has some potential image quality advantages relative to optical scanners. There is not the narrow critical distance that is the basis of all optical FTIR live scan units. Every point in the fingerprint ridge must make actual contact with the optics to be seen. Portions of a ridge, for instance, that are just beyond the critical distance of an FTIR system are invisible. There is no such sharp cutoff in the e-field sensor. Those portions of a ridge that are just out of contact with the sensor surface will be sensed as slightly lighter than the ones that are against it. Contact with the sensor per se is not essential.
The baseline device described above can be the platform for a family of products with increasing degrees of functionality.
The sensor described above is ideally suited to use in situations where there is a powerful processor at hand to perform the minutia extraction. An example is a laptop computer secured by a FingerLoc sensor. The local processor can be used for, say, a fraction of a second to perform minutia extraction.
In other situations there is not ready access to CPU power at the sensor site. A enhanced version of the FingerLoc sensor that includes minutia extraction is aimed at those applications, such as point-of-sale terminals. The extraction is performed with a companion chip in the same package. This companion chip uses more of Harris' DSP experience to rapidly extract the minutia from the image and present a minutia set to the small local control processor for authentication or communication to a central authentication processor.
An even more advanced version allows the authentication to be done in the same package. In this case the extracted minutia are matched against several minutia sets that are stored in local nonvolatile memory. This small database of fileprints can be entered from the sensor or loaded from an external source. When a user touches the device, his fingerprint is processed and the minutia compared with the stored minutia. The result of the search (match or no match) is encrypted and transmitted. The receiver, can decrypt the message, determine the result of the search and take the proper action.
This device will be ideal for self-contained applications where a general purpose computer and reference database are not available. Examples include door access and appliances. This device could also be used to authenticate a user of a smart card, where the authorized user's own minutia set is stored encrypted on the card.
The authors believe that the direct fingerprint reader described in this paper has the potential to help provide inexpensive and reliable personal identification to a wide variety of automated tasks. We hope it will help to foster personal security and privacy in those things that are private, and personal accountability in those that are public.
Here's where you can find out how to get your free copy of Adobe's acrobat reader.