This flexibility arises because the system does away with conventional,
fixed-pattern masks, which are the “stencils” that light passes through to etch
patterns on the surface of the silicon wafer.

The system, outlined in the latest issue of Microelectronic
Engineering (vol 32, p 143), was devised at MIT’s Department of Electrical
Engineering and Computer Science. It is designed to solve two of the most
pressing problems in X-ray lithography— focusing the rays and making and
maintaining a mask.

The X-rays are focused onto the silicon wafer through a computer-controlled
array of hundreds of tiny mechanised windows. Each window has two components.
The upper part is a lens to focus the X-rays, and the lower part is a
micromechanical shutter to turn parts of the beam on and off during
scanning.

The flat lens, known as a zone plate, is similar to a Fresnel lens. It is
patterned with a series of transparent and opaque rings that focus the beam. By
choosing the right X-ray source and designing the zone plates correctly, MIT’s
Henry Smith says that they should be able to focus the X-rays into a spot
measuring just 25 nanometres across.

Each focused section of the beam will then be switched on or off by its own
microshutter. The microshutters will be created using conventional lithography
on a silicon wafer.

Each of the hundreds of windows in the array would be tens or hundreds of
micrometres square. Given that the area to be etched is hundreds or thousands of
times bigger than the focal spot on the silicon, the whole array must be scanned
back and forth in order to fill in the entire area. Using existing X-ray
sensitive materials, Smith calculates that a wafer could be written at a rate of
1 square centimetre per second.

This is no better than the rate at which wafers could be mass-produced
optically ten years ago. But the computer-controlled switches in the new system
allow each new circuit to be different without the need for new masks, while the
higher resolution of X-rays allows each square centimetre of chip to be much
more complex.

The system eliminates the risk of damage to chips that comes with optical
lithography, in which the mask must be held very close to the silicon. It also
compares favourably with electron-beam lithography, which is at least ten times
slower than the new system and has a lower resolution.

The only part of the MIT scheme that may require new technology stems from
the proposed use of uranium to absorb X-rays in the zone plates. Gold rings on
the zone plates would only allow 10 per cent of the X-rays to be used for
patterning. But spent uranium improves the efficiency to 31 per cent. The main
disadvantage with uranium is that it spontaneously combusts in air, but Smith
says that techniques for handling it have been developed in nuclear
laboratories.

The quest to miniaturise silicon components makes X-ray lithography a hot
topic. Other researchers are trying to design reflective lenses—based on
the design of a lobster’s eye—to focus the X-rays (Technology, 6 July, p
18).

The new systems, developed by Annapolis Micro Systems of Annapolis, Maryland,
are based on devices called field programmable gate arrays (FPGAs). Ordinary
gate arrays are chips containing many sets of basic logic gates (And, Or, and
Not). Initially, ordinary gate arrays have no connections between the logic
gates, but they are generally used to make cheap customised chips by hardwiring
the gates together according to the specifications of different customers.

In this way, a single basic chip design can be adapted to perform many
different tasks. The drawback is that once each chip is wired up, the
configuration cannot be changed.

In contrast, FPGAs allow the connections to be changed instantly by software
commands. Until recently they were only used in expensive research and military
systems. The US military, for example, has built a series of reconfigurable
computers called Splash, Splash 2, and WILDFIRE over the past decade.

But when the performance of FPGAs improved, their potential for commercial
computers became apparent. Annapolis Micro Systems believes its systems are the
first commercial products in the field. The development followed a technology
transfer deal with the National Security Agency and the Institute for Defense
Analyses Supercomputing Research Center.

The company has produced two cards, which fit into a standard computer slot
known as the PCI, or Peripheral Component Interface. WILDFORCE is the more
powerful of the two and contains five FPGA chips. Data travels to and from the
host computer to each of four processing chips, while the fifth FPGA controls a
switching network known as the crossbar. The crossbar determines how data flows
between the four other chips.

To use the system, the computing task must first be broken down into its
various parts and the layout for each basic process designed. This involves
choosing combinations of logic gates, known as processors, to perform specific
functions like adding, sorting or filtering inputs. These units must then be
organised so that the links between communicating gates are as short as
possible. The designs for each process, which can eventually be used as designs
for ordinary chips, are then downloaded onto the FPGA chips.

The hardwiring designs can only be downloaded at the beginning of each task.
But the connections between them are determined by the crossbar, which can be
controlled in real time. After the hardware has been set to suit a specific
application, the order in which the logic is performed can be optimised for each
portion of the task.

Just how fast all this turns out to be depends on both the designer’s ability
and the job in hand. The clock speed can vary for every set of hardwiring
designs, and can differ from one FPGA to another.

According to Annapolis Micro Systems’ Bradley Fross, the WILDFORCE board is
particularly useful for rapid prototyping, giving design engineers the
opportunity to experiment more easily with proposed chip configurations.

But the algorithm is more than just a high-tech aid for the removers. Its
first task was to help design bone implants for hip replacements (
International Journal of Robotics Research, vol 15 p 211).

Hip replacements substitute an artificial ball joint for the damaged end of
the femur. This involves drilling a hole in the bone and inserting an implant
with a protruding ball. For the bone and implant to function as one without
using cement—and surgeons prefer not to use cement in half of all hip
replacements—two conditions have to be met.

First, the implant has to lock as tightly as possible into the hole. Second,
the surgeon must be able to insert the implant without putting too much pressure
on the bone.

To fulfil both these criteria, two IBM researchers—Leo Joskowicz, now
at the Hebrew University of Jerusalem, and Russell Taylor, now at Johns Hopkins
University in Baltimore, Maryland—devised a program called Extract. The
program creates a digital model of the implant and the hole, which it uses to
calculate the relationship between a series of points on the implant and the
surface of the bone closest.

The algorithm works by moving the virtual implant a tiny distance out of the
bone and then recalculating the relationship between the points on the implant
and the nearest piece of bone. If the points and surfaces are close or touching,
but not overlapping, then the movement is considered good and the algorithm
moves the implant again.

If the implant and bone overlap, then the model moves the implant back to its
last position and tries a slightly different movement, perhaps adding a small
rotation. This trial-and-error process continues, with backtracking if
necessary, until either the model has built up a continuous path of movements or
the implant gets irrevocably stuck.

Applying the model to 30 test implants, Joskowicz and Taylor were able to
calculate an insertion trajectory with an accuracy of 40 micrometres. With this
level of precision, the algorithm could be used not just to analyse which hip
implants can be inserted and which cannot, but also to control robots involved
in all sorts of complex industrial assembly tasks.

The Swiss “minibot”, which has been under development for two years, measures
about five centimetres across. But though the machine itself is quite large, it
can move in tiny steps of 10 nanometres—one-fiftieth the wavelength of
visible light.

Instead of relying on wheels or conventional mechanical legs to get about,
the minibot shuffles around on legs powered by the reverse piezoelectric
effect—in which electrical pulses are used to deform parts of the
legs.

Each leg is made up of two piezoelectric elements, plus a hemispherical glass
foot making contact with the ground. Controlled by a PC, the voltage across the
piezoelectric components is increased stepwise, and this steadily deforms the
robot’s legs. Friction between the glass foot and the surface keeps the end of
the leg still, so the steel disc that makes up the main body of the robot is
moved forward as the legs deform. When the electrical pulses cease, the leg
snaps back to its original shape quickly enough to overcome friction with the
surface and bring it into line with the body’s new position.

Jean-Marc Breguet, an EPFL researcher, says it is possible to steer the
minibot by varying the voltage applied to the piezoelectric elements. “We can go
in a straight line, we can make a translation, we can make rotations or any
combinations of these,” he says.

The speed of the manoeuvres can also be varied. Using 400-nanometre leg
deformations, the minibot can move at more than 4 millimetres per second and
position itself to an accuracy of 0.5 micrometres. For higher-precision work it
must slow down and move in much smaller steps. The automatic control system uses
laser interferometry to measure the robot’s position accurately and feed the
information back into the PC. So far, this has enabled the researchers to
position the robot with a precision of 10 nanometres in only one direction at a
time. They are developing systems to allow them to achieve this accuracy in
three dimensions at once.

In August, the minibot will begin its first construction job, positioning a
number of rotors with diameters varying between 200 and 500 micrometres. For
this trial, the Swiss researchers will be working with the University of
Karlsruhe in Germany.

The robot will be controlled over the Internet, though for the trial the
control system will be elsewhere in EPFL’s microtechnology lab. “We just want to
show that we’re able to pilot our robot over the Net using vision feedback,”
says Breguet.

For this trial the researchers are aiming for a precision of only 1
micrometre. This is well within the tolerance of the robot, but is at the limits
of the machine-vision controller they will be using for this particular
experiment.

Light can be thought of as a procession of waves, made up of crests and troughs. The “phase angle” indicates which point on the wave cycle between consecutive crests the light has reached in a particular location. Most of the light we see has a random phase, with all the waves out of synch. But with laser light the crests and troughs from all the waves line up, and a “phase mask” can impose an invisible pattern on such beams.

Just like the light and dark regions on a photographic slide, a phase mask has areas of high and low optical density or refractive index. Where the optical density is high, the light slows down and the crests and troughs passing through that point fall out of step, or phase, with their neighbours. The pattern is completely invisible to the naked eye, both on the mask and in the light beam.

Sticking a transparent phase mask over a photo on a credit card would make it very difficult to forge. The authenticity of the card could be checked using a reader made up of a laser light source and a device called an optical correlator. The correlator produces a characteristic interference pattern called a correlation peak if the mask on the card matches a mask held inside the reader. To fool the card reader, a forger would need sophisticated equipment to analyse and replicate the mask.

Bahram Javidi has been leading the research at Storrs, and his latest paper will be published next month in conference proceedings from SPIE, the International Society for Optical Engineering, called Optical Security and Counterfeit Deterrent Techniques (vol 2659).

In more advanced systems, Javidi says, the photograph itself can be coded by running it “backwards” through an optical correlator containing two phase masks. The resulting pattern is an apparently random array of dots, and this is all that is shown on the card. One of the phase masks needed by the optical correlator to decode this image is then stuck on the card, while the other is held inside the card reader.

With this system, it would be impossible for counterfeiters to substitute an alternative photograph using a stolen card alone, not least because they would not know what the original picture looked like and would be unable to work back to analyse the phase masks. It would take billions of years to try all the possible combinations.

Although larger liquid crystal displays have been available in colour for some time, it has been extremely tricky to build in colour on a small scale using conventional LCD techniques. With larger screens, such as those of notebook computers, manufacturers lay down a pattern of red, green and blue gels in front of the liquid crystal. These filter the light and give each subpixel its colour.

This is fine for pixels that are a third of a millimetre wide, but for smaller screens it becomes impossible to align the dots of gel with the micrometre-scale pixels, which are built into a chip behind the liquid crystal.

Now Philip Alvelda, a postgraduate researcher at the Massachusetts Institute of Technology; has managed to build the colour into the chip along with the pixels using standard chip fabrication techniques. This enables him to use pixels that are as little as a few micrometres wide without any colour alignment problems.

The colour is provided by tiny diffraction gratings etched into the chip. The gratings are sets of grooves, reflective at the top and black at the bottom. When illuminated, each grating acts like a prism, dispersing white light into a rainbow-like spectrum.

Although the range of colours in all the spectra are identical, the grating for each colour is designed to send the light out at a slightly different angle, so the red of the first spectrum comes out at the same angle as the green of the second and the blue of the third. The angle of reflection can be adjusted by changing the spacing between the grooves. By using a “window” around the display to mask out all but this part of each spectrum, the subpixels each appear to be the appropriate colour.

Because colour is included at the chip fabrication stage, the colour chips are as easy to produce as the monochrome variety and should be as cheap. Alvelda has now set up the MicroDisplay Corporation in Berkeley, California, and says that he expects to finish the first batch of mass-produced colour displays in early 1997.

The same phenomenon can also create narrow conducting lines within the material. These behave like minute wires, so the research could spawn devices that combine optical and electronic components inside a single, solid substrate.

Holograms can be stored when light causes a local change in the refractive index of the recording material. Most holographic memories rely on the electro-optic effect, in which laser light passing into a crystal sets up a local electric field which changes the material’s refractive index. But these changes are usually small, and the holograms often take a long time to record or demand powerful lasers.

The new technique relies on semiconductors that contain groups of atoms that NEC calls DX centres, which need to capture two extra electrons in order to be stable. At very low temperatures, DX centres normally grab the electrons, giving the material a very high resistance to electric current. But where the material is exposed to laser energy, the electrons can escape. The free electrons form a plasma, which makes the material conductive and changes its refractive index.

Two factors should allow DX-centre semiconductors to store more information than conventional holographic material. First, the conducting lines can theoretically be very narrow – only 100 nanometres across – so the resolution is very high. A trillion points could be recorded in an area of 10 by 10 centimetres or a cubic millimetre, for example.

Secondly, many more plasma-effect holograms can be recorded in the same volume without any interaction between them. Even with conventional materials, a number of holograms can be stored in a single chunk of material by exploiting a technique called angular multiplexing (see “The trillion-bit cube, New ĐÓ°ÉÔ­´´, 13 August 1994). But in photorefractive materials, each hologram can erase part of those recorded before it, because the electric fields may cancel each other out. This limits the capacity of the material.

Plasma-effect holograms can only be erased by heating the semiconductor, which frees all the trapped electrons, and then cooling it again, which re-traps them. If the temperature is kept constantly low, the first recording should be able to carry as much data as the last.

NEC’s early results are promising, but all the experiments have been carried out at temperatures below 70K (−203 °C). For practical systems, DX-centre semiconductors that operate in less extreme conditions must be developed. According to Richard Linke of the NEC centre, the team is already investigating a semiconductor that exhibits some of the properties of a DX-centre material at 250K (−23 °C).