No abstract

Absfrucf-The two-dimensional discrete cosine transform (2-D DCT) has been widely recognized as a key processing unit for image data compressioddecompression. In this paper, the implementation of a 200 MHz 13.3 mm' 8 x 8 2-D DCT macrocell capable of HDTV rates, based on a direct realization of the DCT, and using distributed arithmetic is presented. The macrocell, fabricated using 0.8 /im base-rule CMOS technology and 0.5 pm MOSFET's, performs the DCT processing with I sample-(pixel)-per-clock throughput. The fast speed and small area are achieved by a novel sense-amplifying pipeline flip-flop (SA-FN) circuit technique in combination with t t MOS differential logic. The SA-FN, a class of delay flip-flops, can be used as a differential synchronous sense-amplifier, and can amplify dualrail inputs with swings lower than 100 mV. A 1.6 ns 20 bit carry skip adder used in the DCT macrocell, which was designed by the same scheme, is also described. The adder is 50% faster and 30% smaller than a conventional CMOS carry look ahead adder, which reduces the macrocell size by 15% compared to a conventional CMOS implementation.

No abstract

No abstract

The switch element (SE) is a 622Mb/s, 8 x 8 shared-buffer ATM switch LSI for backbone LAN and WAN applications. The SE has 5Gbps bandwidth, supporting 5 &OS classes delay priority and link-by-link multicast. Up to a 32x32 switch with 2OGbps bandwidth can be configured using multiple SEs and distributor/ arbiter (DA) LSIs. Figure 1 shows the logical queue structure realized in SE. Since the best-effort classes will require much larger per-link buffer than that in SE, the SE can dedicate a small-cell buffer to each class and link, and feed the buffer congestion status signals back to switch access (SA) LSIs that use the expandable per link huge buffer memories. As a result of throughput simulations and chip size consideration, the 64-cell buffer per class, which corresponds to the total buffer size of 320 is selected. Link-by-link multicast function is supported using a shift register type address generator. Figure 2 shows the block diagram of SE. The input cell interfaces establish bit synchronization and cell synchronization, and transform cell data bit-width from 4b to 128b. The routing information is extracted from the cell header. The control logic contains cell counters, threshold registers and comparators. The routing information is transferred to the control logic to decide the cell to be received. The cell data is written to and read from the cell buffer with 128b in parallel sequentially from link0 to link7. Readwrite addresses of the cell buffer are controlled by a shift register type address generator using the information of cell counters and flowcontrol signals received from DA or SA. The control logic generates flow-control signals in time-multiplexed fashion to upstream LSIs using flow-control signals received from the downstream LSIs, and the status of the SE cell counters to avoid the cell buffer overflow. The output cell interfaces transform the bit-width from 128b to 4b. Finally, LVDS output buffers convert the level of output cells from CMOS to LVDS.Each link to or from the SE consists of 5 pairs of LVDS buffers4b data and l b clock. LVDS interfaces have the following advantages: 1) low-power consumption a t high-speed operation, 2) high CMRR and low EMI, 3) low-cost, 4) small-signal distortion, and 5 ) few level conversion ICs through backplanes and PCBs. Figure 3 SE.shows the schematic diagram of LVDS driver and receiver for the The shift-register type address generator shown in Figure 4a has a dedicated structure for readwrite address management of the cell buffer. The shift register is more advantageous in implementing the multicast function than the linked-list because unicast cells and multicast cells can be handled in the same way. Only one clock period is needed to enqueue a multicast cell for any combination of output links. The address generator in the SE has 320 entries, each with 8b for the output link map, l b for the multicast identifier, 3b for the class of delay priority, 9b for the address pointer to the shared buffer and l b for the address pointer parity.Since each data line ...