Systolic array device
A systolic array device according an embodiment includes a plurality of processing units arranged in a matrix form of M by N (M and N are natural numbers). Each of the processing units includes: a processing element configured to perform a predetermined processing based on data received from a processing unit arranged adjacent to one side of the corresponding processing unit to output a result thereof; and a transfer part configured to perform one of an operation of transferring the received data to another processing unit arranged adjacent to the other side of the corresponding processing unit and an operation of transferring the result.
1. A systolic array device comprising:
a plurality of processing units arranged in a matrix form of M by N, where M and N are natural numbers,
a memory storing fault information indicating presence of faults in processing elements of a plurality of processing units; and
wherein each of a first subset of the plurality of processing units includes:
a processing element configured to receive a first input data from one processing unit adjacent to each of the first subset of the plurality of processing units in a first direction and a second input data from another processing unit adjacent to each of the first subset of the plurality of processing units in a second direction different from the first direction, perform a predetermined processing on the second input data using the first input data, and generate a result;
one or more first multiplexers configured to receive the result and the first input data, the one or more first multiplexers configured to selectively output, (i) the result in response to receiving a signal from the memory indicating that a fault is not present in the processing element or (ii) the first input data in response to receiving another signal from the memory indicating that the fault is present in the processing element; and
one or more second multiplexers configured to receive the result and the second input data, the one or more second multiplexers configured to selectively output, (i) the result in response to receiving the signal or (ii) the second input data in response to receiving the other signal;
wherein the plurality of processing units include:
a main processing unit containing a second subset of the plurality of processing units arranged in a matrix form of P by Q, where P is a natural number equal to or smaller than M, Q is a natural number equal to or smaller than N, and a sum of P and Q is smaller than a sum of M and N; and
a preliminary processing unit containing at least one row or column of the processing units added to the second subset of the plurality of processing units, one or more of the each of the plurality of processing units in the preliminary unit replacing one or more of the each of the plurality of processing units in the main processing unit in response to determining that the one or more of the each of the plurality of processing units in the main processing unit are faulty.
2. The systolic array device of claim 1 , wherein the first input data comprises a weight, and a partial sum generated as an output result of the first processing unit, and
wherein the one or more first multiplexers include a first multiplexer for the weight, and a second multiplexer for the partial sum.
3. The systolic array device of claim 1 , wherein the one or more first multiplexers are configured to receive data from one among the plurality of processing units adjacent to each of the first subset of the plurality of processing units in a same column and processing units diagonally adjacent to each of the first subset of the plurality of processing units, and to transfer the received data to another one of the plurality of processing units.
4. The systolic array device of claim 1 , wherein the one or more second multiplexers are configured to receive data from one among the plurality of processing units adjacent to each of the first subset of the plurality of processing units in a same row and processing units diagonally adjacent to each of the first subset of the plurality of processing units, and to transfer the received data to another one of the plurality of processing units.