Memory device and operation method thereof
A memory device and an operation method thereof are provided. The memory device includes: a plurality of page buffers, storing an input data; a plurality of memory planes coupled to the page buffers, based on received addresses of the memory planes, a plurality of weights stored in the memory planes, the memory planes performing bit multiplication on the weights and the input data in the page buffers in parallel to generate a plurality of bit multiplication results in parallel, the bit multiplication results stored back to the page buffers; and at least one accumulation circuit coupled to the page buffers, for performing bit accumulation on the bit multiplication results of the memory planes in parallel or in sequential to generate a multiply-accumulate (MAC) operation result.
1. A memory device including;
a plurality of page buffers, storing an input data;
a plurality of memory planes coupled to the page buffers, a plurality of weights stored in the memory planes, based on received addresses of the memory planes, the memory planes performing bit multiplication on the weights and the input data in the page buffers in parallel to generate a plurality of bit multiplication results in parallel, the bit multiplication results stored back to the page buffers; and
at least one accumulation circuit coupled to the page buffers, for performing bit accumulation on the bit multiplication results of the memory planes in parallel or in sequential to generate a multiply-accumulate (MAC) operation result.
2. The memory device according to claim 1 , further including:
a plurality of conversion units coupled to the memory planes, for converting a plurality of memory cell currents from the memory planes into a plurality of conversion results;
an accumulation detection circuit coupled to the conversion units, for comparing the conversion results of the conversion units with a threshold value, respectively to generate a page buffer selection signal and to generate an accumulation enable signal; and
an output latch, coupled to the accumulation detection circuit, in response to the page buffer selection signal, the output latch selecting corresponding bit multiplication results stored in corresponding page buffers to the at least one accumulation circuit,
in response to the accumulation enable signal, the at least one accumulation circuit sequentially performing bit accumulation operations on the bit multiplication results selected by the output latch to generate the MAC operation result.
3. The memory device according to claim 2 , wherein
when the conversion units are triggered, the memory device performs a hybrid MAC operation; and
when the conversion units are not triggered, the memory device performs a digital MAC operation.
4. The memory device according to claim 2 , wherein
when the conversion units are triggered, the at least one accumulation circuit sequentially performs bit accumulation on at least one target conversion result among the conversion results, the at least one target conversion result being higher than the threshold value; and
when the conversion units are not triggered, the at least one accumulation circuit sequentially performs bit accumulation on all the conversion results, no matter the conversion results are higher than the threshold value or not.
5. The memory device according to claim 1 , further including:
a plurality of conversion units coupled to the memory plane, for converting a plurality of memory cell currents from the memory planes into a plurality of conversion results;
a plurality of accumulation detection circuits coupled to the conversion units, for comparing the conversion results of the conversion units with a threshold value, respectively to generate a plurality of accumulation enable signals; and
an output latch,
wherein the at least one accumulation circuit includes a plurality of accumulation circuits, coupled to the accumulation detection circuits, when enabled by the accumulation enable signals, the accumulation circuits performing bit accumulation operations in parallel on the bit multiplication results from the memory planes to generate a plurality of digital accumulation results; and
the output latch coupled to the accumulation circuits for outputting the digital accumulation results from the accumulation circuits into the MAC operation result.
6. The memory device according to claim 5 , wherein
when the conversion units are triggered, the memory device performs a hybrid MAC operation; and
when the conversion units are not triggered, the memory device performs a digital MAC operation.
7. The memory device according to claim 5 , wherein
when the conversion units are triggered, the accumulation circuits perform bit accumulation in parallel on at least one target conversion result among the conversion results, the at least one target conversion result being higher than the threshold value; and
when the conversion units are not triggered, the accumulation circuits perform bit accumulation in parallel on all the conversion results, no matter the conversion results are higher than the threshold value or not.
8. The memory device according to claim 1 , wherein the at least one accumulation circuit includes
a grouping circuit, for performing grouping operations on the bit multiplication results from the memory planes to generate a plurality of grouping results; and
a counting unit, coupled to the grouping circuit for performing bitwise counting on the grouping results to generate the MAC operation result.
9. The memory device according to claim 1 , wherein
a plurality of bits of each of the input data or each of the weights are divided into a plurality of bit vectors;
each bit of the bit vectors is transformed from a binary format into a unary code;
said each bit of the bit vectors represented in the unary code is duplicated a plurality of times into an unFDP (unfolding dot product) format; and
the multiplication circuit performs multiplication operations on the input data and the weights represented in the unFDP format to generate the bit multiplication results.
10. An operation method for a memory device, the operation method including:
storing an input data in a plurality of page buffers;
based on received addresses of a plurality of memory planes, performing, by the memory planes, bit multiplication in parallel on a plurality of weights stored in the plurality of memory planes and the input data to generate a plurality of bit multiplication results in parallel, the plurality of bit multiplication results being stored back to the page buffers; and
performing bit accumulation on the plurality of bit multiplication results of the memory planes in parallel or in sequential to generate a multiply-accumulate (MAC) operation result.
11. The operation method for memory device according to claim 10 , further including:
converting a plurality of memory cell currents from the memory planes into a plurality of conversion results;
comparing the conversion results with a threshold value, respectively to generate a page buffer selection signal and to generate an accumulation enable signal; and
in response to the page buffer selection signal and the accumulation enable signal, selecting corresponding bit multiplication results stored in corresponding page buffers to sequentially perform bit accumulation operations to generate the MAC operation result.
12. The operation method for memory device according to claim 11 , wherein
the memory device performs a hybrid MAC operation or a digital MAC operation.
13. The operation method for memory device according to claim 11 , wherein
sequentially performing bit accumulation on at least one target conversion result among the conversion results, the at least one target conversion result being higher than the threshold value; or
sequentially performing bit accumulation on all the conversion results, no matter the conversion results are higher than the threshold value or not.
14. The operation method for memory device according to claim 10 , further including:
converting a plurality of memory cell currents from the memory planes into a plurality of conversion results;
comparing the conversion results with a threshold value, respectively to generate a plurality of accumulation enable signals; and
in response to the accumulation enable signals, performing bit accumulation operations in parallel on the bit multiplication results from the memory planes to generate a plurality of digital accumulation results; and
outputting the digital accumulation results into the MAC operation result.
15. The operation method for memory device according to claim 14 , wherein
the memory device performs a hybrid MAC operation or a digital MAC operation.
16. The operation method for memory device according to claim 14 , wherein
performing bit accumulation in parallel on at least one target conversion result among the conversion results, the at least one target conversion result being higher than the threshold value; and
performing bit accumulation in parallel on all the conversion results, no matter the conversion results are higher than the threshold value or not.
17. The operation method for memory device according to claim 10 , wherein
performing grouping operations on the bit multiplication results from the memory planes to generate a plurality of grouping results; and
performing bitwise counting on the grouping results to generate the MAC operation result.
18. The operation method for memory device according to claim 10 , wherein
a plurality of bits of each of the input data or each of the weights are divided into a plurality of bit vectors;
each bit of the bit vectors is transformed from a binary format into a unary code;
said each bit of the bit vectors represented in the unary code is duplicated a plurality of times into an unFDP (unfolding dot product) format; and
a multiplication circuit performs multiplication operations on the input data and the weights represented in the unFDP format to generate the bit multiplication results.