A special feedback technique lets inspection systems ID problems in circuitry far more quickly than with bruteforce methods.
Senior System Engineer
Axsys Technologies Inc.
Rochester Hills, Mich.
Look behind the screen of today's supersized flat-panel displays and you'll find a glass substrate etched with microscopic wiring. No question that this wiring must be error-free; problems would make the whole 2-m display worthless.
A lot of thought has gone into how to inspect this wiring quickly but without missing any errors on the substrate. The usual approach is to mount a video camera and a microscope on a gantry, which moves over the substrate. The motion-control system powering the gantry memorizes the location of several key inspection points. Once over an inspection point, the microscope focuses in on the substrate. Then an image-processing engine examines the video sent back to decide if the circuitry has any errors.
Interestingly, it takes very little time to acquire an image and locate any defects in the field of view. Most of the time that the system spends at an inspection point goes into focusing the lens. That's because the focusing process is iterative.
Specifically, a feedback system finds the point of best focus by maximizing the bandwidth of the image picked up by the video camera. To do so, the system must repeatedly make small changes in focus, then calculate the bandwidth of each image. The bandwidth calculations tell it whether to make another change. Generally, the fastest way to do the job is to start from one end of the focus travel and work through the best focus, then step back to the best focus position. The complete process requires several iterations with their own stops and starts at each inspection position.
Such a brute-force approach wastes a lot of time acquiring and analyzing multiple images. Moreover, each stop-start takes life away from the motor and gears. Designers must drive the system at the highest practical speeds to minimize the time spent traversing the lens through the focus range. That means rapid acceleration as well, inducing large reaction loads on bearings and gear teeth and reducing their lives.
Fortunately there is a fast autofocus method that reduces this time to a fraction of what traditional autofocus schemes require. The key insight is to recognize the usefulness of an analog nulling signal. The signal diminishes as the system approaches focus. And the sign of the signal tells in which direction the focus mechanism needs to go. Use of a nulling signal can eliminate the stops and starts associated with iterative focusing. Inspection throughput rises as well.
It is possible to create such a nulling signal with an auxiliary optical system that creates a shaped beam from a lowpower laser. Microscope optics project this beam onto a bi-cell photosensor.
The illumination comes from a solidstate infrared laser putting out less than 1 mW at 785 nm. A cylindrical lens spreads the beam into a slowly diverging fan shape. Additional aspheric optics give the beam the form of a doubly warped sheet. The part closer to the bicell has negative curvature. The beam has positive curvature farther away. Between the positively and negatively curved portions lies a dividing line where the beam is flat.
The auxiliary optical system is tuned so the laser beam falls on the bi-cell after first traversing the microscope's optical system and then being redirected out of the imaging path by a beam splitter.
The system uses the shaped beam to generate a differential voltage proportional to the distance from best focus. To understand how this works, first consider the case when the system is properly focused. Here, the beam's centerline falls on the bi-cell at the dividing line between the positive and negative sensor elements. If the system is perfectly focused on the substrate pattern, optics project a real image of a single straight line lying along the dividing line between the sensor halves. In this state, equal amounts of infrared radiation fall on the two sensor halves, and their voltage outputs sum to zero.
Now consider the case where the microscope is too close to the substrate under inspection. Then the image's center falls behind the sensor. More light falls on the bi-cell's positive half. The voltages no longer sum to zero and a net positive signal appears at the output.
Finally, suppose the microscope is too far from the substrate. The bi-cell intercepts the beam in its negatively shaped portion. More light falls on the negative bi-cell half, so a net negative voltage appears at the output.
Figures illustrating the sheet's curvature tend to exaggerate the effect for clarity. In fact, only a very slight curvature is needed to generate a strong signal proportional to an out-of-focus condition. The sensor can, in fact, be extremely sensitive to the focus position. It is important to realize that the sensor output is a differential analog signal that responds virtually instantaneously to any change in the focus condition.
THE FEEDBACK LOOP
To synthesize a focusing movement, the system feedback loop first takes the raw output voltage from the sensor and sends it through a feedback amplifier. This amplifier boosts the signal to a useful level and also converts it to a two-bit logic signal.
Specifically, the logic signal can contain an up bit and a down bit. If the down bit is high, then the drive motor moves the microscope toward the substrate. If the up bit is high, it moves the microscope away. If neither bit is high, the drive motor stops. The system does not use the state where both bits are high.
The feedback loop closes through the optical system, which responds instantaneously as the microscope moves. The voltage signal drops as the system approaches perfect focus and reaches zero at perfect focus. In addition, the fast focusing makes it unnecessary to drive the focus mechanism as fast as possible. It can run it at a more controllable speed to help avoid overshooting the in-focus condition.
It is interesting to contrast the operation of the new method with that of traditional focusing. First, thanks to the sign of the sensor-output voltage, the new system always knows in which direction to drive for better focus. This is an improvement over the traditional autofocus method, which requires two measurements with a motion step in between to resolve the direction ambiguity.
Traditional autofocus systems can't start from a nearly infocus condition because their first step could easily carry them past the point of focus. In such a case, the algorithm becomes confused and actually steps away from best focus. To avoid this ambiguous condition, traditional autofocus systems start by driving to the limit in the direction away from the substrate. They can then assume that the direction to best focus is toward the substrate.
The new analog fast-autofocus system does not step, so it can't overstep. Also, it always knows which direction to move, so it isn't easily confused. Consequently, if the system starts close to perfect focus, it will move very little and reach optimum position quickly.
In this way, the new system allows predictive autofocusing. During initial setup, the system goes through an exercise of autofocusing at each inspection point on a test substrate. It remembers the best-focus condition for each point in a calibration table. Later, when it has finished testing at one inspection point during a production test, it can look up the best-focus position for the next inspection point at the same time it looks up the translational coordinates. The system then drives the focus mechanism while slewing the inspection head to the new inspection point.
Thus predictive autofocusing makes much more time available for the initial focus adjustment by piggybacking on the slewing time. The focus drive no longer need be extremely fast. It only must be fast enough to drive from the old focus to the (expected) new focus during the slewing period. Reduced speed means reduced motor output, a drive that can weigh less, less wear on bearings and gears, and longer service life.
This fast analog autofocus system can retrofit onto an existing microscopic inspection system. The necessary components consist of an autofocus head, a Z-axis actuator, controller electronics, and an operator control panel.
The autofocus head carries the photodetector sensor assembly and the laser and optics to generate the doubly warped light sheet. A robust linear-motion actuator carries the entire microscope system, moving it vertically with respect to the substrate under inspection. The control electronics chassis includes the electronics to acquire the sensor output, the computer to make motion decisions and the vertical-axis drive electronics. The operator control panel provides a means for an operator to manually control the system.
We have tested this novel fast-autofocus system in a production environment during the manufacture of large flat-panel displays. It works well with microscope magnifications from 5 through 150 . The hardware includes the infrared laser source and reflective optics to form the shaped beam, the bi-cell sensor, and a microscope body and turret to accommodate them. The video microscope can be built from standard optical components and camera. This whole video-microscope head then mounts on any suitable gantry built for automated inspection applications. The hardware and software for controlling the inspection process and analyzing images is the same as for making the same inspection with a traditional autofocus system.
The difference is that inspection time drops drastically because autofocusing ceases to be a bottleneck.
Axsys Technologies Inc.