Vector Operation Circuit Using Programmable Threshold Switching

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

In computationally intensive tasks, such as artificial intelligence and machine learning applications, vector matrix multipliers require efficient scalar product calculations, but existing circuits face high energy and area requirements due to the need for high-resolution analog-to-digital converters to measure currents from memristor columns.

Innovation Solution

The proposed solution reduces measurement currents by utilizing semiconductor switching elements with programmable threshold voltages, allowing for the use of low-resolution analog-to-digital converters, which reduces energy and space requirements by minimizing current flow through column lines during reduction times.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If high-resolution analog-to-digital converters are used to measure currents from memristor columns, then measurement precision is improved, but energy and area requirements increase

Engineering Contradiction:
Improvecurrent measurement precisionVSAvoidenergy requirement
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent applies preliminary action by reducing measurement currents before they reach the analog-to-digital converter. Current reduction elements are configured to decrease the magnitude of currents flowing through column lines prior to conversion, allowing lower-resolution ADCs to achieve sufficient measurement precision. This preliminary current reduction enables the use of lower-resolution ADCs, thereby reducing energy and area requirements while maintaining adequate measurement capability for vector operations.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If high-resolution analog-to-digital converters are used to measure currents from memristor columns, then measurement precision is improved, but area requirements increase

Engineering Contradiction:
Improvecurrent measurement precisionVSAvoidarea requirement
Core Design Contradiction:
Measurement precisionVSArea of stationary object

Solution Approach 1:

The patent applies preliminary action by reducing measurement currents before they reach the analog-to-digital converter. Current reduction elements are configured to decrease the magnitude of currents flowing through column lines prior to conversion, allowing lower-resolution ADCs to achieve sufficient measurement precision. This preliminary current reduction enables the use of lower-resolution ADCs, thereby reducing area requirements while maintaining adequate measurement capability for vector operations.

Inventive Principle:
Principle #10Preliminary action

3Use of energy by moving object

If measurement currents are reduced by utilizing semiconductor switching elements, then energy and space requirements are reduced, but measurement precision may be compromised

Engineering Contradiction:
Improveenergy consumptionVSAvoidmeasurement accuracy
Core Design Contradiction:
Use of energy by moving objectVSMeasurement precision

Solution Approach 1:

The patent applies preliminary action by reducing measurement currents before they reach the analog-to-digital converter. Current reduction elements are configured to decrease the magnitude of currents flowing through column lines prior to conversion, allowing lower-resolution ADCs to achieve sufficient measurement precision. This preliminary current reduction enables the use of lower-resolution ADCs, thereby reducing energy and area requirements while maintaining adequate measurement capability for vector operations.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent applies parameter changes by modifying the current magnitude parameter through semiconductor switching elements with programmable threshold voltages. These switching elements dynamically adjust current flow parameters, reducing measurement currents to appropriate levels for low-resolution ADCs while maintaining sufficient signal integrity for accurate vector operation results within acceptable error tolerances.

Inventive Principle:
Principle #35Parameter changes

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 efficient vector operations with lower energy and space consumption, suitable for AI and neural network applications, while maintaining accuracy with error tolerance.

Implementation Method 1

semiconductor switching elements with programmable threshold voltages, allowing for the use of low-resolution analog-to-digital converters, which reduces energy and space requirements by minimizing current flow through column lines during reduction times

Methodology Applied
Scientific EffectProgrammable threshold voltage:

Data Source

PatentUS20240265986A1Circuit and method for carrying out a vector operation
Publication Date: 2024.08.08 ROBERT BOSCH GMBH
  • US20240265986A1 patent drawing
  • US20240265986A1 patent drawing
  • US20240265986A1 patent drawing

AI summary

A circuit for carrying out a vector operation. The circuit includes: memory cells, connected to a column line, each having an input terminal and a semiconductor switching element, the gate terminal of which is connected to the input terminal, and the drain or the source terminal being connected to the column line, wherein when a sufficiently high gate voltage is at the gate terminal, the memory cell is activated so that an electrical cell current is conducted out of or into the column line at suitable voltages; an input voltage circuit connected to the input terminals; a detection circuit connected to the column line and having an analog-to-digital converter which determines a digital value corresponding to the intensity of a measurement current flowing into or out of the column line, the detection circuit being configured to detect the digital value determined by the analog-to-digital converter, and reduce the measurement current.