Abstract: In one embodiment, an integrated circuit such as an FPGA includes one or more data I/O blocks, one or more FIFOs, and a FIFO controller. At least one data I/O block receives an incoming bit stream from an external device. At least one FIFO is connected to receive a corresponding incoming bit stream from a corresponding data I/O block. The FIFO controller controls operations of the one or more FIFOs, such that (i) bits from the corresponding data I/O block are written into the at least one FIFO using a FIFO write clock that is based on an incoming clock signal and (ii) bits are read out from the at least one FIFO using a FIFO read clock that is based on a local reference clock signal.

Abstract: In one embodiment of the invention, an integrated circuit, such as an FPGA, comprises a distributed FIFO architecture that supports data transfer from an external device, such as an SDRAM, via an interface that receives a non-continuous, asynchronous strobe clock and a data lane having a plurality of bit lines from the external device. The distributed FIFO architecture comprise a FIFO for each bit line and a FIFO controller. Under control of the FIFO controller, data is written into each FIFO using a FIFO write clock based on the strobe clock, while data is read out from each FIFO using a FIFO read clock based on a local reference clock of the integrated circuit. The distributed FIFO architecture is designed to handle a range of possible phase differences between the FIFO write and read clocks to safely convert from the asynchronous, non-continuous strobe domain to a local continuous clock domain.

Abstract: A multiport memory in one embodiment of the invention includes a memory cell array, where each column in the array has two exterior complementary bitline pairs and zero, one, or more interior complementary bitline pairs. Across each pair of adjacent columns in the array, the adjacent exterior bitline pairs are associated with the same port in the multiport memory. In addition, within each column, the two exterior bitline pairs have the same, non-zero number of crossovers, and, across each pair of adjacent columns, the exterior bitline pairs have different numbers of crossovers. Furthermore, each column has at least one reference signal line located between the two exterior bitline pairs.

Abstract: In certain embodiments, the present invention is a word-line driver for an address decoder that decodes a multi-bit address to enable access to a row of circuit elements such as memory cells in a block of memory implemented in a dedicated memory device or as part of a larger device, such as an FPGA. The word-line driver has a feed-back latch for each word-line that ensures that the word-line is not energized when that word-line is not selected for access. By controlling the feed-back latch using a decoded address bit value rather than a pre-charged enable signal as do some prior-art dynamic word-line drivers, the word-line driver prevents undesirable energizing of multiple word-lines. The word-line driver can be implemented using less layout area and less power than some analogous prior-art static word-line drivers.

Abstract: In one embodiment of the invention, an address decoder for decoding a word-line address to energize a word line in a block of computer memory. Instead of relying on a distinct enable signal, a clock signal provides a reset function and an enable function to the address decoder. In one implementation, the address decoder includes negative-level-sense latches and 3-input AND gates to generate decoded address bits. Using the clock signal as one of the inputs to the AND gates ensures that all of the decoded address bits are 0 when the clock signal is low and that exactly one decoded address bit is 1 when the clock signal is high. In this way, the address decoder ensures that two or more word lines are not energized at the same time.