Abstract: A method and apparatus for processing a bi-directional dataflow are disclosed which permits the transparent movement of data from one processor to another via a shared memory fabric which is connected with both processors. This permits the incoming data of a first processor to be utilized by a second processor thereby freeing that processor from having to handle incoming data. Further, the second processor can handle outgoing data exclusively, freeing the first processor from having to handle outgoing data. In this way, each direction of a bi-directional dataflow may be handled by the maximum capability of a bi-directional capable processing device. The shared memory may comprise a plurality of banks of synchronous dynamic random access memory (SDRAM) devices, and may be used to store packet data in a network.

Abstract: An apparatus is described, comprising a plurality of banks, each bank having a number of columns of non-volatile storage units, each non-volatile storage unit having an input buffer for storing a page of data, the page having an input coupled to the input buffer accepting an input portion of data of a page at a memory speed, the non-volatile storage units storing the data from the input buffer within a memory write time; a plurality of interface buffers; an input bus having an input bus speed which is faster than the memory speed, the input bus being coupled to the plurality of interface buffers; a bus system, connecting each of the plurality of interface buffers to the non-volatile storage units of a column in each of the plurality of banks, supplying data from the plurality of interface buffers to the inputs of the non-volatile storage units at the memory speed.