Input Shaping for Group-Modulated CIM Energy Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Computing-in-memory (CIM) applications face increased energy consumption when processing multi-bit inputs, particularly with group-modulated inputs, which hinders efficiency and accuracy in neural network operations.
Innovation Solution
An input-shaping method and unit that split multi-bit input signals into sub-groups and apply shaping thresholds to increase the probability of bits being zero, reducing energy consumption by optimizing the processing phases and thresholds.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If group-modulated inputs are used to process multi-bit inputs in CIM, then input operation time is reduced, but energy consumption increases
Solution Approach 1:
The patent changes the parameter distribution of input signals by applying shaping thresholds that increase the probability of zero bits. This parameter transformation reduces the number of non-zero input bits that require energy-consuming operations, thereby reducing overall energy consumption while maintaining the fast group-modulated input operation time
Solution Approach 2:
The patent applies partial action by selectively shaping only certain input sub-groups rather than all inputs. By focusing the shaping operation on specific sub-groups that benefit most from zero-bit probability increase, the system achieves energy reduction without unnecessarily processing all input data, thus balancing energy savings with operational speed
2Measurement precision
If input precision in neural network is increased, then processing accuracy is improved, but operating time and power consumption are lengthened
Solution Approach 1:
The patent transforms the parameter distribution of high-precision input signals by increasing the probability of zero bits through shaping thresholds. This parameter change allows the system to maintain high input precision where needed while reducing the number of non-zero operations that consume power, thereby achieving high precision with lower power consumption
3Measurement precision
If input precision in neural network is increased, then processing accuracy is improved, but operating time is lengthened
Solution Approach 1:
The patent applies parameter changes by transforming input signal distributions to increase zero-bit probability. This transformation reduces the number of operations required for high-precision processing, thereby maintaining input precision while reducing the operating time needed to process multi-bit inputs
Data Source
AI summary
An input-shaping method for a group-modulated input scheme in a plurality of computing-in-memory applications is configured to shape a plurality of multi-bit input signals. The input-shaping method for the group-modulated input scheme in the plurality of computing-in-memory applications includes performing an input splitting step, a threshold setting step and an input shaping step. The input splitting step includes splitting the multi-bit input signals into a plurality of input sub-groups via an input-shaping unit. The threshold setting step includes setting at least one shaping threshold via the input-shaping unit. The input shaping step includes shaping at least one of the input sub-groups according to the at least one shaping threshold via the input-shaping unit to form a plurality of shaped multi-bit input signals so as to increase a probability of a bit equal to 0 occurring in the at least one of the input sub-groups.