Transparent Display Electromagnetic Computing for Lower AI Latency

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing artificial intelligence systems face performance degradation due to high communication latencies and data volumes during model parallelism, particularly in tasks involving complex operations like inference and training, which are exacerbated by the need for all-to-all communication among processing units.

Innovation Solution

Utilizing electromagnetic energy to perform collective operations by displaying neuron activations on transparent displays and detecting these signals with sensors, allowing for passive computation and reduced n-to-1 data transfer, thereby minimizing the need for traditional n-to-n data transfer.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If model parallelism is used to accelerate AI operations, then computation speed is improved, but communication latency and data transfer volume increase

Engineering Contradiction:
Improvecomputation speedVSAvoidcommunication latency
Core Design Contradiction:
ProductivityVSLoss of time

Solution Approach 1:

The patent replaces traditional electrical/optical communication systems with electromagnetic radiation-based communication. Processing units emit electromagnetic signals that propagate through space to reach other processing units, eliminating the need for physical communication channels and reducing communication latency in distributed AI systems.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent introduces spatial dimensionality by allowing electromagnetic signals to travel through three-dimensional space between processing units. This enables direct line-of-sight communication paths that bypass traditional hierarchical communication structures, reducing the number of communication hops and associated latency.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Adaptability or versatility

If all-to-all communication is implemented among processing units, then model parallelism functionality is achieved, but energy consumption and memory load increase

Engineering Contradiction:
Improvemodel parallelism functionalityVSAvoidenergy consumption
Core Design Contradiction:
Adaptability or versatilityVSUse of energy by moving object

Solution Approach 1:

The patent implements self-service through passive electromagnetic signal propagation. Once electromagnetic signals are emitted by processing units, they automatically propagate through space and are detected by receiving units without requiring active management, routing, or buffering infrastructure, significantly reducing energy consumption compared to traditional active communication systems.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent extracts the communication function from traditional electrical/optical channels and implements it through electromagnetic radiation in free space. This separation eliminates the energy overhead associated with maintaining physical communication infrastructure and enables more efficient all-to-all communication patterns.

Inventive Principle:
Principle #2Taking out (Extraction)

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 computational expense and improves efficiency, power consumption, and latency in AI model training and inference by leveraging the innate properties of electromagnetic signals for collective computation.

Implementation Method 1

each emitter is configured to display an electromagnetic signal... an electromagnetic sensor configured to detect a collective electromagnetic signal from the plurality of emitters

Methodology Applied
Scientific EffectElectromagnetic radiation: Electromagnetic Induction

Implementation Method 2

detect a collective electromagnetic signal... based on a collective property of the electromagnetic signals

Methodology Applied
Scientific EffectWave superposition: Interference

Data Source

PatentEP4636533A1Performing computing operations using a collective property of electromagnetic energy observed through transparent displays
Publication Date: 2025.10.22 MICROSOFT TECHNOLOGY LICENSING LLC
  • EP4636533A1 patent drawingFigure 1
  • EP4636533A1 patent drawingFigure 2
  • EP4636533A1 patent drawingFigure 3

AI summary

A computing system includes a plurality of processing units and a plurality of emitters. Each emitter is coupled to at least one of the plurality of processing units and is configured to display an electromagnetic signal at a location of the emitter based on instructions from an associated processing unit. The computing system further includes an electromagnetic sensor configured to detect a collective electromagnetic signal from the plurality of emitters.