Bend Light, Store Bytes

February 2001
By Henry S. Kenyon

Researchers explore alternative recording systems using light and microscopic dots and dashes.

From clay tablets to magnetic tape, civilizations have found ways to store important information; however, the silicon revolution has led to an overabundance of data. While existing electronic media have kept pace with this demand, new technologies could offer massive storage coupled with fast retrieval.

Scientists are exploring new systems that could become commercially available in the coming decades while continuing work on improving the capabilities of existing media. Some of these techniques, such as light-based memory, may open new paths for high-speed computing, experts say.

IBM’s Almaden Research Center in San Jose, California, has a long tradition of developing data storage technologies. According to Currie Munce, Almaden’s director of storage systems and technology, though IBM is actively researching promising techniques such as holographic storage, magnetic hard disks will continue to be the storage medium of choice for the foreseeable future. This situation has much to do with economics, he explains. Magnetic hard disk technology is very mature and is supported by a broad infrastructure of suppliers and manufacturers. More than $2 billion a year is spent on research and development, doubling magnetic storage capacity every year. Hard disk drives also are inexpensive and ubiquitous, with more than 200 million units shipped annually.

New technologies face a number of hurdles when competing against magnetic hard disk drives. Munce says it is difficult to amass the large amounts of research and development funding and create an infrastructure of suppliers for a new system, and a new device also must be inexpensive and operate at performance levels that are high enough to compete with hard disk drives. “This is not to say that eventually something might not become competitive, but it will be difficult [for the new technology], and it will have to possess some attractive or unique attributes. It might also move into a niche market because it has some special performance attribute desired in that domain where it may offset costs or other attributes of hard disk drives,” he says.

Magnetic storage development will continue at its current pace for at least three more years, Munce predicts. Technical issues revolving around the concentration of data storage will then begin to slow development to a pace comparable to that of the semiconductor industry, but the medium will advance rapidly for another five to seven years, he says. In that time, magnetic recording will reach capacities of at least 400 to 500 gigabits per square inch. At this density, a single 3.5-inch disk platter on a desktop computer’s hard drive would possess something on the order of 500 to 750 gigabytes per platter. “You can fit up to five platters in a 1-inch-high desktop drive. If you were creative, in five to seven years you could envision 2.5 to 3 terabytes in a desktop drive,” he contends.

Munce believes that hard disk drive capacity probably will not reach these levels for several reasons. The demand for storage capacity is slow, only growing at 40 to 50 percent per year. Another issue is performance. Hard disk drives have only one actuator moving the head that reads and writes the data. With information more densely packed on a disk, access capacity is limited because it takes longer for the actuator to locate the data. This is leading to a disparity between how quickly data can be accessed and retrieved and how much can actually be stored, he says.

However, researchers are attempting to address these issues in the server market, where hard disk size has shrunk. This allows the disks to operate at more rotations per minute so the actuators can seek information more quickly. Another trend is to run drives and array systems in parallel to enhance performance, Munce reports. These changes are occurring in the server market, but not in the desktop arena because PC prices are traditionally driven by the least expensive part in a single device instead of multiple devices, he says.

A recent development in storage technology is a memory system for laptop computers and other portable devices. Magnetic random access memory (MRAM) is chip-based and works like static random access memory and dynamic random access memory (DRAM) to store data; however, MRAM does not lose data when the power is shut off.

The device’s memory is stored in a manner similar to flash memory, Munce explains. Instead of trapping a charge and capacitance in a silicon semiconductor type device, a magnetic film is used to store the charge and its orientation. By measuring the change of magnetic resistance in the film, data can be identified as a one or a zero. He notes that this shares many of the attributes of magnetic recording, especially a recording head, though the device’s wiring and processing are more similar to a DRAM system.

MRAM devices will be able to store more information, access it more quickly and use less power than devices such as hard disk drives. Developers expect it to enter the commercial market in 2004.

Another long-term research project with potential is holographic storage. IBM has been working with a consortium of companies and the Defense Advanced Research Projects Agency to develop the technology. One of holography’s advantages is that large amounts of information can potentially be archived and accessed very quickly. Unlike magnetic or optical recording systems that store information in a nanometer-thin layer on top of a spinning disk, holographic media record data throughout the volume of a cube that can be millimeters thick. “In essence, you use not only the surface area but also the depth of a material to store information,” Munce says.

The technology is also inherently parallel because images are used to write data into the cube. A two-dimensional (2-D) image is used to write the data. The hologram is created from interfering beams, but the image and format that a user reads is 2-D.

Parallelism allows greater access speeds. In recent demonstrations, data was read out of a holographic system at 10 gigabits per second. Magnetic systems relying on serial techniques typically read out at speeds of one gigabit per second, Munce says.

Holography also enables users to retrieve and store information by manipulating beams of photons instead of using mechanical devices. Such a process would retrieve information in microseconds as opposed to milliseconds because the beams have no mass, Munce says. Two beams are used to create a hologram, an object beam and a reference beam. Once an image is created and stored, it can be played back through the reference beam. The hologram is stored inside the recording device. Only the reference beam is used to retrieve the data. Holographic storage is also a reciprocal process. “If I put the original hologram back into the cube storing the information, I’ll actually replay the reference beam that created it [the hologram],” he says.

Multiple holograms can be stored in the same media by multiplexing, or slightly changing the angle of the reference beam. This method allows for very high data densities, with a 1,000- or 10,000-hologram storage capacity in a data cube. Units of holographic information are measured in megapixels, representing a million bits of data. To play each hologram back, the user views the reference beam that created a specific image. This type of retrieval is different from the traditional address-based system. Holograms instead are located by the picture, individual word or data that best matches a logic string. Rather than using processors and memory to do all of the data comparisons to find information, holography moves photons around to read information, Munce explains.

One possible application for holographic memory could be in very large supercomputers used for data-intensive work. It might be possible to install a holographic memory cache that provides microsecond-fast access times. In an average computer, the cache is next to the processor, followed by the DRAM and the hard drive. The cache is close to the access point so that it does not have to encounter bus speed limitations to access the DRAM. But access speed decreases when data moves from DRAM to the hard disk drive. A holographic cache in the 2- to 10-gigabyte range placed between the DRAM and the disk drive provides a more uniform step down in speed, he says.

Holographic data storage is still in the research and development phase, and its uses are limited to niche applications. While the technology has demonstrated a write-once read-many-times function, it is still missing the right storage media, Munce says. The ideal material would have photosensitivity, speed, a long lifetime and the optical qualities to make holography a more viable technology. The current medium of choice for holography is lithium niobate. It is iron-doped, inorganic and possesses very low photosensitivity.

The lack of photosensitivity is a problem because it takes seconds to write information and it is possible to accidentally erase or overwrite data when it is accessed. “You have to come up with various fixing schemes to be able to somehow write it and be able to fix it like developing film. This way, when you later do something to it, it won’t overwrite or erase the information that is there,” he explains.

If the right material can be found, it will probably be an organic photopolymer. The trouble with these photopolymers is that when they are exposed, they shrink and experience other effects that distort the film, Munce notes. The resulting optical aberrations limit the density at which the information can be stored. Finding techniques to prevent data loss from overwriting is another issue of concern, he says. IBM is working with photochemistry and photopolymer firms to develop a solution, he says.

The company also is investigating a new technology called Millipede that could have applications in mobile communications. Munce describes the device as “an elegant digital phonograph” consisting of a series of small cantilever styluses embedded in a silicon wafer. Resting on a smooth moving plastic surface, the tips of the devices are heated briefly with an electric current to leave a slight indentation in the polymer. The tip can then detect these pits and read them as data. The pits also can be erased and written over by reheating the tips. The marks made by the tips are about 40 nanometers wide. This creates the potential to store hundreds of gigabits per square inch, a task that would be difficult to do with magnetic media today, Munce notes.

The prototype consists of 1,024 styluses arranged in a 32 x 32 array in an area that is 3 x 3 millimeters. One reason the device uses so many parallel styluses is that it is a mechanical system with a lower data rate. The parallel tip arrangement provides for higher data and throughput speeds.

While Millipede does not have the data density of magnetic storage, it does have more than DRAM or solid-state memory. It is a silicon micromachine, but it is able to store data at levels exceeding magnetics. This system offers advantages in wireless computing and communications devices. Applications include cellular telephones and other portable communications devices. “Everyone’s going to carry around a PDA [personal digital assistant] or a gigabyte in their cell phone five years from now,” he says.

Munce notes that in coming years, devices such as cellular telephones will have built-in features such as audio recording, a digital camera and a scanner to record text. Devices in military use will have maps loaded into them as well as other applications for data, logistics and tactical combat applications such as head-up displays. All of this requires gigabytes of information. Millipede provides this data storage more cost effectively than does a hard drive. It also uses less power, which is a critical issue for battery operated devices in the field.

The device is more rugged than a hard disk drive because Millipede does not have bearings and moving parts. “You could drop this thing, bounce it off walls and not damage it. You can’t completely say that about hard disk drives today,” he says. However, he cautions that Millipede is still in the exploratory research stage. He believes that it will be three to four years before this technology reaches the commercial marketplace.