SESSION SA21

SALON 10-15

MEMORY: NV AND EMBEDDED

Chair: T-S. Jung, Samsung Ltd., Kyungki-do, Korea
Associate Chair: T-S. Feng, United Microelectronics, Taiwan

21.1 - Fully-Parallel 25MHz 2.5Mb CAM - 8:30 AM

K. Schultz, F. Shafai, R. Gibson, A. Bluschke, D. Somppi

Nortel Semiconductors, Ottawa, Ontario, Canada

A 25MHz 64kx40 CAM, the largest fully-parallel CAM to date, employs NAND match line chains for low power, and two chains per word for speed. A word-sliced architecture is augmented by operation-specific self-timing loops. The 230mm2 7.5W chip in 5-metal 0.35um CMOS, incorporates BIST and is cascadable.

21.2 - A 1M-cell 6b/Cell Analog Flash Memory for Digital Storage - 9:00 AM

P. L. Rolandi, R. Canegallo, E. Chioffi, D. Gerna, G. Guaitini, C. Issartel, A. Kramer, F. Lhermet, M. Pasotti

SGS-Thomson Microelectronics, Agrate Brianza, (MI), Italy

A 1M-cell memory prototype for non-volatile digital storage achieves up to 6b per cell (equivalent of 6Mb). The chip, suitable for applications such as digital still camera, stores 24VGA full-color JPEG images or 40 minutes of compressed speech. The chip uses standard 3V 0.5um Flash-EEPROM process with 2-poly 2-metal. Array density is 257Mb/cm2.

21.3 - 400-Level, Constant Programming Rate Single-Pulse Programmable Nonvolatile Analog Memory - 9:30 AM

K-h. Kim, K. Lee

Korea Advanced Institute of Science and Technology, Taejeon, Korea

An EEPROM programing scheme is based on a nonvolatile analog memory (NVAM). Constant programming rate with single program pulse drastically enhances programming speed and accuracy. A test chip containing 8x128 NVAM cells (9x13.6um2 cell) is fabricated using 0.8um CMOS. Each cell stores 8b in 360us.

BREAK 10:00 AM

21.4 - A 3.3V 133MHz 32Mb Synchronous Mask ROM - 10:15 AM

J-H. Park, D-W. Lee, H-S. Im, J. Lee, Y-H. Lim, W-K. Lee, E-D. Kim, W-M. Lee, K-D. Suh

Samsung Electronics, Ltd., Kiheung, Korea

A 32Mb mask ROM has SDRAM-compatible interface and control. Using time-multiplexed page sensing and dynamic load with mirror bias sense amplifier, 133MHz operation is obtained with single-metal 0.4um CMOS. The die is 62.45mm2. The typical data access time is 5.8ns.

21.5 - A 33GB/s, 13.4Mb Integrated Graphics Accelerator and Frame Buffer - 10:45 AM

R. Torrance, I. Mes, B. Hold, D. Jones, J. Crepeau, P. DeMone, D. MacDonald, C. O'Connel, P. Gillingham, R. White1, S. Duggins2, D. Fielder2

MOSAID Technologies,Inc., Carp, Ontario, Canada

1Accelerix Inc., Ottawa, Ontario, Canada

2Symbionics Ltd., Cambridge, England, UK

A complete PC graphics system including the DRAM frame buffer, implemented in a 0.35um/0.5um merged DRAM/logic process, is integrated on a 129mm2 die. The frame buffer includes 13.4Mb DRAM, with 4kb wide I/O port, 3 4kb-wide serial output registers, and 4kb-wide SIMD processor that allows 4kb wide graphics rastor operations. Registers and processor are pitch-matched to the DRAM, allowing 33GB/s bandwidth between pixel processor and DRAM.

21.6 - Compression/Decompression DRAM for Unified Memory Systems: a 16Mb, 200MHz, 90% to 50% Graphics-Bandwidth Reduction Prototype - 11:15 AM

Y. Yabe, Y. Aimoto, M. Motomura, T. Takizawa, T. Miyamoto1, T. Iwasaki1, Y. Nakazawa, T. Fujii, M. Hamada, N. Nagai, M. Yamashina

NEC Corp., Sagamihara, Japan

1NEC Informatec Systems., Ltd., Kanagawa, Japan

A CompressDRAM integrates compression/decompression hardware into a packet-oriented DRAM. By embedding this chip in a unified memory system, graphics-related memory bus traffic is reduced from 90% to 50%. A 16Mb prototype integrates a 200MHz compressor/decompressor with 800MB/s Synclink-type DRAM I/F using a 0.35um merged logic-DRAM process.

21.7 - A 128Mb Early Prototype for Gigascale Single-Electron Memories - 11:45 AM

K. Yano, T. Ishii, T. Sano1, T. Mine, F. Murai, T. Kure, K. Seki Hitachi Central Research Labs., Kokubunji, Tokyo, Japan

1Hitachi Device Engineering, Kokubunji, Tokyo, Japan

Large-scale integration of the single-electron memory targets minimum bit-cost using double-stacked 2F2/bit-cell. The cell-to-cell variations of characteristics that make large-scale integration difficult are compensated for by dummy-cell-referenced verified read/write. The cell is 0.145um2/b using 0.25um technology.

CONCLUSION


Go back to Table of Contents