Note: This is one in a series of blogs about Fingerprint on Card technology. To view the full series, click here. Part of this blog was written by NXP’s partner in biometric authentication, Precise Biometrics.
Biometric payment cards, equipped with fingerprint on card technology, will perform authentication using your fingerprint. The sensor integrated in the card will scan your fingerprint while you put it in the POS terminal slot or you tap it on the terminal display. Biometric algorithms will then extract data from the scan (the so called templates) and compare it to the reference template stored in the secure element of the card. If the comparison yields a positive match, the card holder will be authenticated and the purchase will be completed.
For Fingerprint on Card technology to succeed, and gain widespread acceptance, it will have to blend seamlessly into the payment process, creating a positive customer experience. The technology will need to work quickly and accurately, without compromising security, and using only existing infrastructure. High performance means delivering speed, power, and accuracy.
The Implementation Matters
How the Fingerprint on Card technology is implemented has a direct influence on card performance. In particular, the configuration of hardware and software interactions, along with the algorithms used to extract and match data, are vital aspects to optimize performance.
That’s why, when it came to designing a fingerprint on card solution, NXP partnered with Precise Biometrics, a company that specializes in biometric technology for secure identity authentication. Their high-performance solutions for biometric authentication are used hundreds of millions of times every day, worldwide.
The Precise Biometrics Approach
Working closely with NXP, Precise Biometrics developed a built-for-purpose algorithm that ensures NXP’s fingerprint on card technology excels in terms of speed, power, and accuracy. NXP asked Precise Biometrics to explain how they optimized these three key parameters. Here’s what they had to say:
1. Speed = Contactless Fingerprint Verification Time within 1s
It’s important to keep execution times to a minimum, since fingerprint verification adds three extra steps to the payment transaction:
To streamline the payment process, and prevent unwanted delays at contact and contactless terminals, the general guideline is to have all three steps of fingerprint verification in under one second which comes on top to the standard payment transaction time.
Given the amount of data associated with a fingerprint image, and the limited computing capabilities of a EMV card which is only powered by the POS terminal without the support of a battery or super capacitor, the key success factor of the fingerprint algorithm is the ability to split the fingerprint extraction and fingerprint matching into two separate processes.
The extraction process, which is more compute-intensive but does not need highest security protection, executes in the card’s microcontroller, a low-power IC specially chosen for the task. This lets the extraction process make use of the microcontroller’s accelerated computing capabilities and reduces the time needed to complete certain mathematical operations.
The matching process, which is less compute-intensive, executes in the secure environment of the secure element. From a security perspective the usage of a secure element for matching is required, since the match process and final results are kept hidden in a secure area of silicon.
In combination with a sensor which is quick at capturing and transferring the image, the target to process the entire fingerprint verification within 1s can be ensured (more on that in an upcoming blog).
Dividing tasks, by placing extraction in the microcontroller and matching in the secure element, makes optimal use of the hardware resources in NXP’s design and creates a solution that delivers both speed and security.
2. Efficiency = Low-Power Operation, Without a Battery or Super Capacitor
To reduce the cost of manufacturing and deployment, the Fingerprint on Card format needs to work with existing terminal infrastructure.
Today’s terminals are designed for lower-power operation and are specified at a minimum field strength of just 1.5A/m. In other words, not only does the fingerprint authentication process need to happen within 1s, it needs to meet this speed requirements while operating on a very limited power budget.
Some Fingerprint on Card systems may be able to work with available field strength, but don’t always achieve the 1s target. Also, some fingerprint on card systems add a battery or super capacitor to the design to increase the power budget. This adds cost and can impact the robustness of the design. Super capacitors are particularly tricky, since they need a certain amount of time before being able to power the card system, and this increases transaction time.
An optimized sensor which is very power efficient in combination with an optimized extraction and matching process ensures that the fingerprint on card solution runs by using the existing field strength. All in all, the right system set up including the right components and implementation enabled a low-power operation while meeting the speed requirement of 1s.
3. Accuracy = A Balanced FAR/FRR Tradeoff
There is always a tradeoff, when configuring the match parameters to be used with biometric authentication, between security and convenience. Focus only on security, and define parameters so tightly that it’s difficult to get a positive match, and you risk compromising convenience, since the system is more likely to refuse an authorized user, even if the user tries repeatedly to verify their fingerprint. On the other hand, put too much emphasis on convenience, and define parameters too loosely, so it’s easier to get a positive match, and you risk compromising security, since the system may allow fraudsters to gain authorization.
To balance this tradeoff, and create a system that delivers acceptable levels of both security and convenience, developers perform tests involving thousands of attempts to access the system. They then use two metrics to track how often the system gets it right. The first metric, the False Rejection Rate (FRR), indicates how often the system wrongly rejects an authorized user, and the second metric, the False Acceptance Rate (FAR), indicates how often the system wrongly accepts an unauthorized user.
Comparing the FRR and FAR results makes it easier to find the area where there’s an acceptable balance between security and convenience. As shown in the graphic, the parameters that yield results between T1 and T2 offer a reasonable tradeoff.
To create a payment card with fingerprint on card functionality that delivers accuracy high enough to satisfy the strict requirements of payment transactions, developers typically aim to produce a FAR/FRR tradeoff that is at least as good as present-day payment cards that use a 4-digit PIN. To reach this level of security, the biometric performance of a fingerprint on card function is trimmed and optimized to reach 3% FRR at a FAR of 1 in 10,000. With this FRR/FAR trade-off, one false fingerprint in 10,000 attempts is, in theory, successfully accepted and three in every 100 genuine fingerprints are falsely rejected.
The FAR/FRR trade-off is complicated by the fact that, like every fingerprint, like all biometric data, varies due to age, gender, origin, profession, and other factors. With fingerprint on card technology, one of the biggest challenges in this regard is the size of the sensor and, as a result, the amount of data captured with each image capture. A large sensor, sized big enough to capture every aspect of the fingerprint on even the biggest of fingers, is cost prohibitive, and requires more power than is readily available on a contactless card. These cost and power constraints mean the Fingerprint on Card function has to make due with a sensor that is much smaller than the average finger. To address this, the Precise Biometrics algorithm makes use of multiple image samples collected from the sensor. The algorithm then sorts, combines, and optimizes different image samples to produce the most efficient and most secure matching process.
To optimize the statistical FRR and FAR settings of the system, Precise Biometrics depends on large databases of fingerprints. These databases are carefully collected to take into account a variety of external factors, such as temperature, humidity, gender, age, and more. Complex machine learning methods are then used to train and trim the algorithm to meet the required system performance in all expected user scenarios.
The Bottom Line: Better Performance
The NXP solution for Fingerprint on Card functionality, based on NXP´s Secure Processing Module, leverages algorithms from Precise Biometrics to enhance performance in terms of speed, power, and accuracy.
Listen to Our Podcast
To learn more about the biometric payment cards, its challenges, and its next steps, listen to the NXP and Precise Biometric podcast on the subject, recorded live at Money 20/20 Europe on June 4.
Next Up: Manufacturing
The next topic in this series is production. We’ll look at what’s needed to add fingerprint on card technology to the manufacturing process and how existing techniques can be used to ensure scalability relating to industrialization.
The next blog is scheduled to post in two weeks.