Abstract: A memory sense amplifier for a static random access memory includes a pair of transistor amplifiers respective to the bit lines threading the memory. The power consumed is minimized without sacrificing speed of operation by temporarily connecting the source electrodes of the transistor amplifiers to the bit lines to allow them to track the states of the bit lines before a current path is completed to the drains of the transistors to allow them to draw current from the bit lines, thereby minimizing the time that the sense amplifiers are permitted to draw current from the bit lines. In addition, an economy of circuitry is achieved by eliminating the need for a separate latch circuit by disconnecting the sense amplifiers from the bit lines and thereafter enabling them to latch the information state read from the bit lines.

Abstract: The single-polysilicon active pixel comprises a photo site located on a substrate for generating and storing charge carriers, the charge carriers being generated from photonic energy incident upon the photo site and semiconductor substrate, a photo gate, a transfer transistor and output and reset electronics. The gate of the transfer transistor and the photo gate are defined in a single layer of polysilicon disposed on the semiconductor substrate. The source of transfer transistor is a doped region of substrate, referred to as a coupling diffusion, which provides the electrical coupling between the photo gate and the transfer transistor. The coupling diffusion allows for the transfer of a signal stored in a photo site under the photo gate to the output electronics for processing. The single-polysilicon active pixel may be operated by biasing the transfer transistor to the low operating voltage of the pixel, for example, 0 volts.

Abstract: The imaging pixel according to the present invention includes a floating gate pixel node capable of nondestructive readout and active source follower output circuitry suitable for combination with other like imaging pixels to form an imaging array.

Abstract: Efficient decoding of an hierarchical, variable length, encoded data sequence containing embedded uncoded data into a sequence of fixed length instructions for subsequent processing by a digital video processor or the like is realized in an apparatus including a decoder having a plurality of variable length code decoding elements and a control structure embedded within each decoding element for transferring the decoding operation to an appropriate one of the decoding elements in response to a prior output from the decoder element. As the encoded data sequence is processed by the apparatus, a predetermined length of the sequence is stored in a register. The control structure further responds to the encoded data sequence to initiate selection of either the predetermined length of the sequence stored in the register or a portion of the decoder output as the fixed length instruction to be output by the apparatus.

Abstract: In known speech recognition systems, processors and methods, utterances are analyzed to obtain a set of reference signals. An unknown signal may be compared with the reference signals. The unknown signal may be said to be the reference signal with which it most closely corresponds as defined by some correspondence measure. Known signal recognition arrangements using multiple processor cells tend to be expensive, in part because they tend to use many processor cells.The disclosed system, processor and method contemplate an arrangement including an array of processor cells for time warping an unknown signal having m elements with respect to a reference signal having n elements or vice versa. The cells, responsive to control signals on a control diagonal, generate the correspondence measure. As the signals propagate through the array, the instant arrangement recirculates signals from cells near one (or first) periphery of the arrangement, e.g.

Abstract: To substantially increase the bandwidth of a random access memory (RAM), a shift register is disposed within the memory array such that the shift register lies parallel to the word lines and is connected to at least individual ones of the bit lines contained within the array. Separate high-speed serial input and output lines are provided by the shift register. These lines are in addition to and operate independently of the slower speed input and output lines normally provided by the RAM. Through this arrangement, a row of data can be transferred to and from the memory array at a rate substantially faster than the single-bit access rate of the RAM.

Abstract: Known signal processors for matching signal patterns commonly compare an unknown signal with one of a set of reference signals. Various comparison techniques are known. One comparison technique for solving a parenthesization problem includes an orthogonal array of interconnected cells which are adapted for dynamic programming and for extending data and control information in a generally left-to-right direction as well as in a bottom-to-top direction. For solving a pattern matching problem, known arrangements for extending control information in a generally left-to-right or bottom-to-top direction do not appear to be satisfactory. The disclosed signal processor for matching signal patterns and for dynamically time warping an unknown input signal with a reference input signal generates a measure of the correspondence between the input signals. In generating the correspondence measure, the processor includes an arrangement for controlling all processor cells on a predetermined diagonal of the array of cells.