How do you think the new GigE standards will influence the machine vision industry?
Respond or ask your question now!
Information is as much a weapon as ordnance. Small wonder, then, that defense imaging is an active area of development. The technology has gone far beyond the short-wave IR (1 to 2 µm) night-vision goggles of yesteryear. Today, cutting-edge systems are leveraging materials like amorphous silicon and carbon nanotubes to image over the mid-wave IR (MWIR, 3 to 5 µm) and long-wave IR (7 to 14 µm) spectral bands.
Traditionally, LWIR systems have used mercury-cadmium telluride (HgCdTe) detectors. The downside is that the material is expensive, not to mention environmentally toxic; more important, for LWIR operation, HgCdTe must be cooled to liquid nitrogen temperatures. That may not present a problem for applications like airborne hyperspectral imaging, but for man-portable battlefield imagers, a Stirling cooler adds weight, bulk, and an additional point of failure.
The benefit of HgCdTe is performance, particularly noise-equivalent temperature difference (NETD), which used to be almost an order of magnitude better than that of microbolometers. Recent advancements to thermal detectors have more than cut their NETD in half, however, making bolometers increasingly effective alternatives for a range of defense applications. Gone are the days of low-resolution, cantilever microbolometers on 100-µm pixel pitches. Now, the focus is on more efficient, and ever smaller, suspended pixels. At L-3 Communications Infrared Products (Dallas, Texas), for example, engineers have developed an amorphous silicon microbolometer array with a 17-µm pixel pitch, a factor of two smaller in area than the previous standard of 25 µm.
Microbolometer arrays operate uncooled and are more economical than HgCdTe. To further reduce cost and simplify fabrication, the L-3 group uses amorphous silicon rather than the more common vanadium oxide. Amorphous silicon has demonstrated its effectiveness as a high-performance uncooled IR sensor—the L-3 devices achieve NETDs of better than 50 mK.
As in all things, there are tradeoffs. The reduced area might facilitate larger arrays, and hence, better spatial resolution, but smaller pixels mean a smaller signal collection area. This, says L-3 chief technologist Charles Hanson, brings up one of the benefits of the thermal approach. "On a thermal detector, as opposed to a photon detector like HgCdTe, you can get that signal back by increasing the thermal isolation," he notes. "The steady-state signal is inversely proportional to the conductance, so if you reduce the thermal conductance, you end up with the same responsivity as the larger pixels but with the advantages of a smaller detector."
The relationship between conductance and signal strength adds several degrees of freedom to the design process. "Shrinking the pixel gives you more trade space," says Hanson. "You get a constant performance, which means the optics can be smaller, lighter, and cheaper. You can fit more detectors on a wafer so the detectors are also cheaper. You also get an additional degree of flexibility in that if your optics are smaller, you can afford to increase the size of your aperture and get more sensitivity."
We live in the real world, so there are challenges to accompany the benefits. A microbolometer converts thermal differentials into current. To thermally isolate the pixels, they are suspended at each end by narrow legs. The smaller the pixel, the thinner the legs, which makes the assemblies susceptible to thermally induced stresses. Addressing those stresses requires a delicate balancing actually between material thicknesses, surface treatments, and annealing processes. More recently, L-3 has taken its 17-µm pitch designs to VGA (640 × 480) to XGA (1024 × 768) resolution. Applications include imagers for weapons sights or armored vehicles.
Connecting the Dots
Advancements continue in photonic LWIR sensors, as well. At the Department Of Electrical and Computer Engineering at the University of Massachusetts, Lowell, associate professor Xuejun Lu and his group have been developing 320 × 256 pixel focal plane arrays of quantum dot infrared photodetectors (QDIPs) based on indium arsenide quantum-dots in indium arsenide/gallium arsenide (InAs/GaAs) material systems, which is about 2.5 orders of magnitude cheaper than HgCdTe. Due to the three-dimensional quantum confinement of carriers in quantum dots (QDs), the QDIP technology offers promising advantages for low-level IR detection, including intrinsic sensitivity to normal incident radiation and long excited-state lifetime. These characteristics permit efficient collection of photo-excited carriers and ultimately lead to high photoconductive gain, quantum efficiency, and photoresponsivity. The normal-incidence detection capability greatly simplifies the fabrication for large format (1000 × 1000 pixel) focal plane arrays.
One of the challenges with even modestly sized focal plane arrays is the sheer number of pixels that need to be interrogated. To avoid bottlenecks, Lu's group has been working toward on-chip processing with high-speed electronics that would separate out static pixels from those that change from frame to frame in a given video stream. By simply imposing an XOR logical process to send only data from pixels that have new information, the scheme dramatically reduces traffic over the communications bus, reducing latency.
Of course, on-chip processing requires on-chip circuitry, which presents something of a challenge. GaAs has a very different thermal expansion coefficient and lattice constant than silicon, making it incompatible with conventional integrated-circuit fabrication processes. Rather than wrestling with flip chip bonding or moving circuitry from a substrate to the focal plane array, Lu's group is pursuing an entirely different approach—carbon nanotubes (CNTs).
With ultra-high carrier mobility, CNTs can operate at speeds in excess of 5.6 GHz, showing tremendous promise for electronics applications. In the case of Lu's project, they hold two other significant advantages. First, they are IR transparent, which means they can be deposited directly on top of the focal plane array without compromising performance. "In the midwave IR or longwave IR, the transmission is very high," says Lu. "We tested this and got over 90 percent transmission through the carbon nanotube layers."
Better yet, circuitry based on the nanotubes doesn't require high-temperature processing—or, indeed, photolithography at all. The CNTs can be dissolved in solution and printed onto the substrate. Because they act as a network of connections, they don't even need to be oriented. "The goal is that after we make the focal plane array, we don't have to do anything, just basically fit it into the printer and start printing without using any photolithography tools." Lu and his group have already successfully printed circuitry onto polymer substrates. Now, they are working to print the circuitry directly on top of their QDIP arrays, as well as developing the electronics to allow the user to toggle between MWIR and LWIR operation.