Multiply-Accumulate Circuit With State-Dependent Charging Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Analog systems used for multiply-accumulate operations in neural network models face challenges in achieving high accuracy and power efficiency, particularly in improving the resolution and accuracy of the multiply-accumulate operation.
Innovation Solution
A multiply-accumulate operation device with input lines, multiplication units, an accumulation unit, a charging unit, and an output unit, where pulse signals with specific pulse widths are input, and the charging speed and threshold values are set based on the accumulation state to enhance the resolution and accuracy of the operation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If analog system is used for multiply-accumulate operation, then power consumption is reduced, but operation accuracy is insufficient
Solution Approach 1:
The patent applies dynamics by making the charging speed variable rather than fixed. The charging unit changes the charging speed based on the accumulation state of the accumulation unit, allowing the system to adaptively optimize both power consumption and accuracy. This dynamic adjustment resolves the contradiction by enabling the system to operate efficiently while maintaining high precision through state-dependent parameter modulation.
Solution Approach 2:
The patent implements parameter changes by modifying the charging speed parameter according to the accumulation state. By changing this physical parameter dynamically, the system achieves both low power consumption and high operation accuracy. The threshold value is also adjusted based on accumulation state, further enhancing the precision of the multiply-accumulate operation while maintaining energy efficiency.
2Device complexity
If fixed charging speed is used, then device complexity is reduced, but resolution and accuracy are insufficient
Solution Approach 1:
The patent applies self-service by enabling the charging unit to automatically adjust its charging speed based on the accumulation state feedback. The system uses its own internal state information to regulate its operation, eliminating the need for external complex control mechanisms. This self-regulating approach achieves high resolution and accuracy without proportionally increasing device complexity.
Solution Approach 2:
The patent implements feedback by using the accumulation state as a control input for adjusting the charging speed. The accumulation unit's state is continuously monitored and fed back to the charging unit, which modifies its charging behavior accordingly. This feedback loop enables precise control and high resolution in the multiply-accumulate operation while keeping the control architecture relatively simple.
3Adaptability or versatility
If threshold determination is performed after charging starts, then calculation timing flexibility is improved, but timing control complexity increases
Solution Approach 1:
The patent applies preliminary action by performing threshold determination after charging has already started, rather than waiting for charging completion. This allows the system to determine the multiply-accumulate result during the charging process itself, providing timing flexibility without requiring complex post-charging processing. The output unit can generate results at optimal moments during the charging phase.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The proposed solution improves the accuracy and resolution of the multiply-accumulate operation by optimizing the charging speed and threshold values, allowing for more precise calculation of the multiply-accumulate results in analog systems.
Implementation Method 1
The accumulation unit accumulates a sum of charges corresponding to multiplication values generated by the plurality of multiplication units
Implementation Method 2
The charging unit charges, at a charging speed associated with an accumulation state of the accumulation unit, the accumulation unit in which a sum of charges corresponding to multiplication values have been accumulated
Implementation Method 3
The output unit outputs a multiply-accumulate signal representing a sum of multiplication values by executing threshold determination using a threshold value associated with the accumulation state of the accumulation unit on a voltage held by the accumulation unit
Data Source
AI summary
A multiply-accumulate operation device, circuit and method are disclosed. In on example, a multiply-accumulate operation device includes input lines, multiplication units, an accumulation unit, a charging unit, and an output unit. Pulse signals having pulse widths corresponding to input values are input to the input lines. The multiplication units generate, based on the pulse signals, charges corresponding to multiplication values obtained by multiplying the input values by weight values. The accumulation unit accumulates a sum of the charges corresponding to the multiplication values. The charging unit charges the accumulation unit at a charging speed associated with its accumulation state. The output unit outputs a multiply-accumulate signal representing a sum of the multiplication values by executing threshold determination using a threshold value associated with the accumulation state of the accumulation unit on a voltage held by the accumulation unit after the charging by the charging unit is started.