Multiply-Accumulate Unit Parasitic Capacitor Bias
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current multiply-accumulate calculation devices and neuromorphic devices face complexity in generating a negative bias for neural network arithmetic operations, leading to intricate circuit configurations.
Innovation Solution
The device employs a plurality of variable resistance elements with parasitic capacitors to generate output signals by multiplying input signals with weights, and an accumulate calculation unit sums these signals after transient responses from charging and discharging have stabilized, allowing for a wide range of bias adjustments from positive to negative values without complicating the circuit.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional circuit configurations are used to generate negative bias, then the bias can be adjusted in a wide range, but the circuit configuration becomes complicated
Solution Approach 1:
The patent extracts the bias generation function from the main calculation circuit by utilizing the transient responses of parasitic capacitors in the multiple calculation elements. Instead of adding separate bias generation circuits, the invention uses the inherent parasitic capacitance already present in the calculation elements, thereby avoiding circuit complexity while achieving wide-range bias adjustment including negative values.
Solution Approach 2:
The multiple calculation elements serve dual purposes: they perform the primary calculation function (multiplying input signals by weights) and simultaneously generate the required bias through their parasitic capacitor transient responses. This self-service mechanism eliminates the need for additional dedicated bias generation circuits, resolving the contradiction between bias adjustment capability and circuit complexity.
2Adaptability or versatility
If the calculation period includes transient responses from parasitic capacitors, then negative bias can be generated, but the calculation timing becomes more complex
Solution Approach 1:
The patent applies preliminary action by intentionally designing the calculation period to include the transient response phase of the parasitic capacitors. Rather than trying to eliminate or separately manage these transient responses, the invention incorporates them into the calculation period from the outset, allowing the bias generation to occur naturally during the expected operation window. This approach simplifies timing control by aligning the bias generation with the existing calculation timeline.
Solution Approach 2:
The patent converts the potentially harmful effect of transient responses (which could cause calculation errors) into a beneficial feature for generating negative bias. By strategically including the transient response period within the calculation period, the invention transforms what was previously a source of operational complexity into a useful mechanism for achieving wide-range bias adjustment, thereby resolving the contradiction between bias polarity range and timing control ease.
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
This approach enables the adjustment of biases across a wide range without adding complexity to the circuit configuration, effectively generating negative biases and allowing for efficient neural network operations.
Implementation Method 1
transient responses caused by charging to parasitic capacitors of the plurality of first multiple calculation elements
Implementation Method 2
each of the plurality of first multiple calculation elements is a variable resistance element
Implementation Method 3
each of the plurality of first multiple calculation elements is a tunnel magnetoresistance effect element
Implementation Method 4
calculate a sum of the first output signals output from the plurality of first multiple calculation elements in a calculation period
Data Source
AI summary
A multiply-accumulate calculation device includes: a plurality of first multiple calculation elements configured to generate first output signals by multiplying a first input signal corresponding to an input value by a weight and output the first output signals; and an accumulate calculation unit configured to calculate a sum of the first output signals output from the plurality of first multiple calculation elements in a calculation period from a point in time at which transition to a steady state has occurred after transient responses caused by charging to parasitic capacitors of the plurality of first multiple calculation elements according to input of the first input signal to a point in time after transient responses caused by discharging from the parasitic capacitors of the plurality of first multiple calculation elements according to input of the first input signal have started to be generated.