SANTA CRUZ, Calif. Behind the spectacular panoramic images beamed back from Mars in recent weeks is a full-custom CCD imaging chip designed to meet the stringent demands and environmental conditions of the Martian surface. The 1,024 x 2,048-pixel chip, which handles imaging for each of the nine cameras on both the Spirit and Opportunity rovers, mandated a rethinking of CCD design.
The frame-transfer charge-coupled device takes in photons, converts them to electrons and then converts them back into the voltage domain for processing by camera electronics. But the Mars rover cameras aren't ordinary digital cameras, as Mark Wadsworth, the former Jet Propulsion Laboratory (JPL) principal engineer who designed the chip, found out.
The first challenge was that all nine cameras on board each rover had to use identical CCD imaging chips to restrain costs. In addition to the two science "pancams" (panoramic cameras) that beam back color photos from Mars, these include four engineering "hazcams" (hazard avoidance cameras), two "navcams" (navigation cameras) and one microscopic imager.
"The specs were all over the map, and we needed one device to meet them all," Wadsworth said.
Factor in the radiation from the space voyage and the Martian surface, the daily temperature swings from --90C to +20C, the need for extreme photosensitivity and the limited bandwidth in Mars-to-Earth communications, and the result is an imaging chip that's very different from what you find in a commercial digital camera.
"We were trying to come up with ways to push CCDs a little further than they've been pushed before, to improve performance, lower the power and lower the peripheral circuitry needed to operate them," said Wadsworth, who today heads a design startup called Tangent Technologies.
An off-the-shelf CCD camera chip wouldn't work, he said, so the design was done at JPL. "The device had to be incredibly low-noise, very sensitive optically and able to run extremely fast," Wadsworth said. "You can pick any two of these three and meet them with something off the shelf, but to find all three required a custom design."
All of the requirements had to be met because of the functions of the different cameras. With the pancams, Wadsworth said, photosensitivity is key. "If there's a photon out there, the science guys want to know about it," he said. But with the navcams and hazcams, the priority is getting fast images so the rovers don't crash into things as they move about.
All of the rover cameras are, in effect, 1-megapixel monochrome digital cameras. The colors in the pancam images come from a sophisticated set of color filter wheels, and the sweeping panoramas are mosaics of many individual images.
Why only 1 Mpixel, when your digital camera at home probably has 3 or 5?
"It's a matter of what you do with the images after you take them," Wadsworth said. "It takes a finite amount of time to get images out of the rover, link them and transfer them."
Indeed, the rovers' direct-to-Earth data transfer rate is a mere 12 kbits/second, whereas communication with the Mars orbiters goes up to 128 kbits/s.
Further, Wadsworth noted, NASA is extremely conservative when it comes to technology. "They demand that something have a heritage and some number of years of application somewhere before they allow it anywhere near a mission," he said.
The CCD chip includes a 1,024 x 1,024 imaging area and a 1,024 x 1,024 image storage area. At three gates per pixel, it yields a little over 6 million gates. It can transfer an acquired image to the frame storage area in about 20 milliseconds and read out images at 200 kpixels/s. The chip is fabricated in a 3-micron CCD process using three levels of polysilicon and two levels of metal.
A charge-to-voltage conversion circuit demanded particular design attention. "Everyone calls it an amplifier, but it's basically an electrometer that measures electrons and converts them to voltage," said Wadsworth. "We get a conversion of about 5 microvolts of signal for every electron that's put on the sensor node."
The chip took a good deal of design work, Wadsworth said, to provide low noise, low-light-level image fidelity and an ability to handle very large signals. In the lab, he noted, the imager can detect signals of as few as four photons; in the camera, minimal signal detection is probably around 20 photons, far lower than commercial digital cameras.
The CCD chip also has extremely large "storage wells" behind each pixel. The chip can store a little more than 200,000 electrons in each well, Wadsworth said, compared with perhaps 20,000 for a commercial digital camera. Larger well sizes help the cameras cope with radiation and also with the extreme temperature range, since thermally generated background charge can fill the wells.
Ironically, it's not the cold of the night but the "heat" of the Martian day that produces this background charge along with the long exposure times required to form an image. "The colder you go, the better fidelity you have with the image," Wadsworth said. "The devices have some cooling built in, but they're really at the mercy of what's going on temperature-wise on the surface of Mars."
The large well capacity didn't jibe with submicron design rules, Wadsworth noted, which is one reason for the 3-micron CCD process. "We have lots of room, so using looser design rules allowed us to get to where we wanted to be and build in extra margin for yield," he said.
The chip is an analog, full-custom design, constructed using a layout editor and Spice simulator from Tanner EDA. "The layout of the imaging area was fairly straightforward and would probably bore most designers," commented Wadsworth, who said it took him four weeks overall to design the CCD. "The real design challenge was in the electrometer circuit."
Wadsworth noted that transistors on CCDs are made in buried-channel technology and cannot be turned off completely. "I had to generate [Spice] models that reproduced that behavior, and come up with a substantial amount of noise modeling," he said. "I created as simple a circuit as I could that produced not only very high sensitivity, in terms of microvolts per electron, but was also very linear over a large number of electrons."
The circuit, said Wadsworth, demonstrates about 99 percent linearity over a signal size of 450,000 electrons comparable to a 2.5- to 3-volt signal swing. Wadsworth said that Tanner EDA, which provides low-cost custom-design tools, fit the bill for this design mission.
"One thing I've learned is that if you don't have to use a complicated layout tool, don't do it," he said. "The nice thing about the Tanner tools is that when you buy them, they're usable, unlike some of the larger packages, where you have to define basic components and models."
Yet, there's room for improvement. "I hope somebody, someday, comes out with a good automated layout tool for analog," Wadsworth said. "I haven't seen it."