Introduction

We have developed a system [Potter et al., 1999], called Leginon, for automatically acquiring large numbers of high quality transmission electron micrographs. The automated system proceeds by acquiring a low magnification image, identifying potential features of interest in this low magnification image, locating the feature of interest at the center of the field of view, automatically focusing at high magnification under low dose conditions, and finally acquiring the high magnification image. The system can acquire up to 1000 high magnification images per day.

The Leginon system has been implemented on a Philips CM200 Transmission Electron Microscope equipped with a Gatan MSC CCD camera. The application is built on a multi-layered software library which provides an environment for portable and extensible microscope control and image acquisition in a distributed software environment [Kisseberth et al., 1998]. The software library provides control of all of the essential microscope and camera parameters necessary to achieve automated image acquisition.

One of critical steps in implementing the automated data acquisition system requires that target locations be identified in a low magnification image and then accurately repositioned to the center of the viewing area for subsequent high magnification imaging. There is typically a two order of magnitude difference between the low (660x) and high (38,000x) magnification images. As a result the effects in the digital image of any inaccuracies in locating the specimen at low magnification are scaled up by a factor of 100 in the high magnification image. Centering an identified target location for subsequent high magnification imaging typically requires moving the specimen by many thousands of nm and accurately locating the target to within a few hundred nm.

For example, in figure 1 is a low magnification [660x] image of a vitreous ice specimen prepared over a Quantifoil grid [Ermantraut et al., 1998]. The goal of the automated system is to locate the holes containing ice of the correct thickness at the center of the screen and subsequently to obtain a high magnification image of the specimen within the hole. Holes which have been identified as being of suitable thickness are indicated on the figure. Each hole is approximately 2000nm in diameter and our imaging conditions require that the center of the hole be located to within 200nm of the center of the field of view. Relocating a hole which is near the periphery of the low magnification image to the center of the imaging area requires moving the specimen over distances of up to 10,000nm. Specimen movements of this size cannot be achieved using the image shift coils, which would be very accurate, and instead must be achieved using the microscope goniometer, which is a mechanical device. This places stringent requirements on the accuracy of the goniometer controlling the specimen movement.

	Accurate relocation	1 mm error in relocation
	Field of view at 38,000x on a 1Kx1K digital camera.
Figure 1: Low magnification (660x) image of a vitreous ice specimen prepared over a Quantifoil grid is shown on the left. Indicated targets must be relocated to the center of the imaging area for subsequent high magnification image acquisition. Accurate relocation of features to the center of the field of view is required for automated data acquisition. Our own experience and experimental measurements show that in practice the errors associated with relocating a specimen can be as much as 1000nm.

The Philips CM200 that we are using is equipped with a computer controlled compustage [de Jong, et al., 1993]. The specifications provided by the manufacturer of this device state that the reproducibility of the goniometer is better than 100nm and our own measurements confirm this. However the issue in specimen relocation is not one of reproducibility, as this requires only that we are able to return precisely to a known goniometer position and does not require that displacements along the goniometer axes be precisely related to real world displacements relative to the specimen. Our own experience and experimental measurements show that in practice the errors associated with relocating a specimen can be as much as an order of magnitude worse than the reproducibility specifications, i.e. a target area could more than 1000nm off center after relocation. The targeting accuracy may be improved by a series of refinement steps in which the target location in the original image is compared against the currently located center and a new goniometer position calculated. However this refinement step is time consuming and requires that the magnification be switched back to the low magnification setting every time a new target is to be imaged. Continuously switching between magnification settings in turn contributes to shifts in the beam and other instabilities in the microscope performance.

Our practical goal was thus to improve the targeting accuracy of the goniometer by at least a factor of 5. We have achieved this by characterizing the performance of the compustage, modeling its behavior and using this model to calculate the goniometer movements required for target relocation. This resulted in a tenfold improvement in the accuracy of the goniomter. The improvement in the accuracy is sufficient to achieve target centering without the need for additional refinement methods.

Page 2 of 7