Memory lookup computing mechanisms
AssigneeSamsung Electronics Co., Ltd.
InventorsPeng Gu, Krishna T. Malladi, Hongzhong Zheng
Life of this patent
Estimating the term…
Abstract
According to some example embodiments of the present disclosure, in a method for a memory lookup mechanism in a high-bandwidth memory system, the method includes: using a memory die to conduct a multiplication operation using a lookup table (LUT) methodology by accessing a LUT, which includes floating point operation results, stored on the memory die; sending, by the memory die, a result of the multiplication operation to a logic die including a processor and a buffer; and conducting, by the logic die, a matrix multiplication operation using computation units.
Claims20
A method for a memory lookup mechanism in a high-bandwidth memory system, the method comprising:
- using a memory die to conduct a multiplication operation using a lookup table (LUT) methodology by accessing a LUT, which comprises floating point operation results, stored on the memory die;
- sending, by the memory die, a result of the multiplication operation to a logic die comprising a processor and a buffer;
- and
- conducting, by the logic die, a matrix multiplication operation using computation units.
A high-bandwidth memory lookup mechanism system comprising:
- a logic die comprising a processor and a buffer, wherein the processor conducts a matrix multiplication operation using computational units;
- and
- a memory die comprising a lookup table (LUT), wherein the LUT comprises floating point operation results.
A memory die comprising a lookup table (LUT) stored thereon, and configured to perform:
- a floating point full product multiplication method comprising:
- using, by the memory die, a whole scalar floating point input to retrieve a whole scalar multiplication result from the LUT, the LUT comprising whole product multiplication results;
- sending, by a buffer, the whole scalar multiplication result to an accumulator;
- and
- accumulating, by the accumulator, the whole scalar multiplication result;
- a floating point partial product multiplication method comprising:
- using, by the memory die, a fraction scalar floating point input to retrieve a partial product multiplication result from the LUT, the LUT comprising partial product multiplication results;
- sending, by the memory die, the partial product multiplication result to the buffer;
- sending, by the buffer, the partial product multiplication result to an adjustor;
- accumulating, by the accumulator, the partial product multiplication result;
- a floating point tensor operation method comprising:
- using, by the memory die, a tensor floating point input to retrieve a tensor multiplication result from the LUT, the LUT comprising tensor multiplication results;
- sending, by the memory die, the tensor multiplication result to the buffer;
- sending, by the buffer, the tensor multiplication result to a tensor post-processor;
- and
- a floating point fused layer operation method comprising:
- using, by the memory die, a whole scalar floating point to retrieve a whole scalar multiplication result from the LUT, which comprises whole product multiplication results;
- sending, by the memory die, the whole product multiplication results to a buffer die comprising a plurality of buffers;
- sending, by the buffer die, the whole product multiplication result to the buffer;
- and
- sending, by the buffer, the whole product multiplication result to a tensor processor.
Dependencies are inferred from the claim text.