From the Labs: Information Technology

New publications, experiments and breakthroughs in information technology–and what they mean.

Kate Greenearchive page

April 21, 2009

Improved Memory
New algorithms double flash capacity without shrinking transistor size.

**Compact memory:** A micrograph shows the two 32-gigabit sections of a new flash memory chip.

Source: “A 5.6 MB/s 64 Gb 4b/Cell NAND Flash Memory in 43nm CMOS”
Cuong Trinh et al.
2009 IEEE International Solid-State Circuits Conference, February 10, 2009, San Francisco, CA

Results: Researchers at Toshiba and SanDisk, a maker of flash memory devices in Milpitas, CA, have built a 64-gigabit chip that holds four bits of data per memory cell, twice as much as the cells in conventional chips.

Why it matters: To increase the amount of data that can be stored in memory chips, engineers typically shrink the transistors that make up the individual memory cells. However, as transistors get smaller, their reliability tends to decrease because they generate more heat and leak more electrical current. While SanDisk researchers are still exploring ways to make transistors smaller without compromising reliability, the new approach makes it possible to store more data without shrinking transistors.

Methods: In conventional flash memory, a transistor stores two bits of data, each defined by a distinct voltage level. A variation of the technology can store four bits per transistor, but this requires more finely tuned voltage levels that can be disrupted by extreme voltage differences between transistors, effectively erasing the data. The SanDisk researchers wrote an algorithm that controls the way data is written to the chip so that the voltage differences between neighboring transistors are kept to a minimum.

Next steps: The company expects its chips to go into production within the first half of 2009. Future chips may use a similar principle, but since the electrical characteristics of transistors will change as smaller ones are developed, applying this approach to new generations of memory chips will require new algorithms that take these changes into account.

Smarter E-Mail
Prototype draws data from multiple sources to figure out where to send messages

Source: “Semantic Email Addressing: The Semantic Web Killer App?”
Michael Genesereth et al.
IEEE Internet Computing 13 (1): 48-55

Results: Researchers at Stanford University have developed a prototype system that allows people to send e-mail to individuals or groups without knowing the recipients’ e-mail addresses or names. The system scours databases and websites to find addresses for people who fit selected criteria, then sends messages to those addresses.

Why it matters: The system would make it easier for e-mail to reach the correct destination, even when the sender isn’t sure what that destination is. It would allow a person to send a message to “Bob Jones,” say, even if Bob Jones changed jobs and has a new address–or to direct an e-mail to “head of the marketing department,” even if the person holding that position has changed. It could also be used to find several people at once who fit certain criteria.

Methods: The researchers developed software that analyzes requests, such as “Professors who went to Harvard before 1970.” Since data needed to fulfill this request will come from various databases that categorize information in different ways (“professors,” “Harvard,” and so on), the researchers also devised an intermediary program. That program translates these requests into terms that will be understood by the databases being queried. For example, a request for Harvard might be directed to a number of database categories, such as “college,” “institute,” and “university.”

Next steps: The system, which currently gleans addresses from a specific set of university and research-institute databases, must be adapted to work with more databases. It can also be used to gather e-mails from websites that use a standard framework for categorizing information. And the system could be used to filter spam. For example, someone could choose to accept e-mail only from “researchers studying the semantic Web,” and the system would identify addresses of those people and allow messages from them.

Keep Reading

Large language models can do jaw-dropping things. But nobody knows exactly why.

And that's a problem. Figuring it out is one of the biggest scientific puzzles of our time and a crucial step towards controlling more powerful future models.