Matrix Circuit Vector Mapping Without Analog-to-Digital Converters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vector matrix multipliers in neural networks require significant space and energy due to the use of analog-to-digital converters for current intensity determination, which are inefficient and costly.
Innovation Solution
A method using time periods instead of current intensity to determine the product sum in a matrix circuit, eliminating the need for analog-to-digital converters by employing memory cells with semiconductor switching elements and adjustable threshold voltages, and measuring elapsed time to calculate output vector components.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If analog-to-digital converters are used to determine current intensity in matrix circuits, then measurement precision is improved, but device complexity and energy consumption increase
Solution Approach 1:
The patent extracts and eliminates the analog-to-digital converter from the matrix circuit by using a different measurement approach. Instead of measuring current intensity directly with complex ADC circuits, the invention measures the time period during which current flows through the memristors, thereby removing the need for ADCs and reducing device complexity while maintaining measurement capability.
Solution Approach 2:
The patent substitutes the electronic measurement system (current intensity measurement with ADCs) with a temporal measurement system (time period measurement). By measuring how long current flows rather than how strong the current is, the invention replaces complex electronic conversion circuits with simpler timing circuits, reducing both device complexity and energy consumption.
2Measurement precision
If analog-to-digital converters are used to determine current intensity, then measurement precision is improved, but energy consumption increases
Solution Approach 1:
The patent removes the energy-intensive analog-to-digital converter from the system by alternative measurement methodology. The time period measurement approach eliminates the need for high-power ADC circuits, thereby significantly reducing energy consumption while still providing accurate measurement of the scalar product result.
Solution Approach 2:
The invention substitutes the high-energy electronic current measurement system with a low-energy temporal measurement system. By using timing circuits instead of ADCs, the patent dramatically reduces energy consumption during the measurement phase while maintaining the ability to accurately determine the scalar product of input vectors and memristor conductivities.
3Productivity
If memristors are arranged in large matrix circuits (100-1000 rows and columns), then productivity is improved, but device complexity increases
Solution Approach 1:
The patent substitutes complex high-resolution measurement circuits with simpler temporal measurement circuits. This substitution allows large-scale matrix circuits to be implemented with reduced complexity per element, as the timing measurement approach requires simpler readout circuitry compared to ADC-based current measurement, thereby enabling scalable deployment of large neural network accelerators.
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 reduces energy consumption and space requirements, enabling efficient and scalable vector operations without the need for analog-to-digital converters.
Implementation Method 1
Each memory cell is set up to conduct an electric current from the first line to the third line depending on the memory state and the voltages applied to the first, second and third lines
Implementation Method 2
Each memory cell has a semiconductor switching element with a control terminal that is connected to the second line assigned to the column in which the semiconductor switching element is located. The semiconductor switching elements can have an adjustable threshold voltage, wherein the memory state of a memory cell depends on the respective threshold voltage
Data Source
AI summary
The disclosure relates to a method for mapping an input vector to an output vector by means of a matrix circuit which has memory cells arranged in a matrix in a plurality of rows and a plurality of columns and first, second and third lines, each memory cell having an adjustable memory state, is connected to the first line (22) of the corresponding row, is connected to the second and third lines of the corresponding column and is set up to generate an electrical current (I1, I2, I3) depending on the memory state and voltages applied to the first, second and third lines, is connected to the second and third lines of the corresponding column and is arranged to conduct an electric current (I1, I2, I3) into the third line (26) as a function of the memory state and voltages applied to the first, second and third lines, each memory cell having a semiconductor switching element (28) with a control terminal which is connected to the second line (24) of the corresponding column; wherein input voltages (U1, U2, U3) corresponding to components of the input vector are applied (110) to the first lines; wherein for each column: a ramp voltage (V1, V2, V3) is applied (120) to the second line assigned to the column, the level of which is increased with time (130); a total current is detected at the third line assigned to the column and a time period elapsed since a start time of the level increase of the corresponding ramp voltage is determined (150) until the magnitude of the total current reaches a certain current magnitude threshold (Ig) (140); and a component of the output vector corresponding to the column is determined (170) based on the elapsed time period (t1, t2, t3).