Low power hardware architecture for handling accumulation overflows in a convolution operation

A method, comprising:

• in response to a running total computed by a first accumulator of a first convolver unit satisfying one of at least one overflow condition, disabling a dot product computation module of the first convolver unit and setting an output of the first convolver unit equal to a predetermined value, wherein disabling the dot product computation module comprises one or more of clock gating the dot product computation module or clamping one or more inputs received by the dot product computation module to a zero value;
• performing an accumulation operation by a second accumulator of a second convolver unit while the dot product computation module of the first convolver unit is disabled;
• upon completing the accumulation operation by the second accumulator of the second convolver unit, setting an output of the second convolver unit equal to an accumulated value;
• and
• at a time instance following the completion of the accumulation operation by the second accumulator at the second convolver unit, sampling the accumulated value from the output of the second convolver unit and sampling the predetermined value from the output of the first convolver unit.