Fenced Memory Regions for Secure Computer Vision Processing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing computing architectures face challenges in ensuring user privacy and security while allowing untrusted applications to access computer vision data stored in trusted memory regions, leading to potential privacy breaches and reduced processing throughput.

Innovation Solution

Implementing a fenced memory region (FMR) controlled by a computer vision processor to manage access to a trusted memory region (TMR), where access is granted or denied based on FMR requests and responses, ensuring secure communication through a data fabric and security processor, and allowing access based on user consent.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If untrusted applications are allowed to access computer vision data in trusted memory regions, then processing throughput is improved, but user privacy and security are compromised

Engineering Contradiction:
Improveprocessing throughputVSAvoiduser privacy and security
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The memory space is segmented into a trusted memory region (TMR) and a fenced memory region (FMR). The FMR acts as an isolated buffer that separates untrusted application access from the secure TMR, allowing controlled data flow while maintaining security boundaries. This segmentation enables both high-throughput processing and security by preventing direct access to sensitive regions.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The fenced memory region serves as an intermediary buffer between untrusted applications and the trusted memory region. Data is first placed in the FMR by trusted applications, then untrusted applications can access it without compromising the TMR. This mediator approach enables throughput improvement while maintaining security through controlled access pathways.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Productivity

If direct access to trusted memory region is allowed for applications, then processing efficiency is improved, but unauthorized access and privacy breaches occur

Engineering Contradiction:
Improveprocessing efficiencyVSAvoidunauthorized access and privacy breaches
Core Design Contradiction:
ProductivityVSObject-affected harmful factors

Solution Approach 1:

Data is pre-positioned in the fenced memory region by trusted applications before untrusted applications need to access it. This preliminary action in the FMR prepares the data in a secure buffer zone, allowing efficient retrieval by untrusted applications without requiring them to access the TMR directly, thus preventing unauthorized access while maintaining efficiency.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The memory system implements local quality by creating a specialized fenced memory region with specific access characteristics. The FMR has different access permissions and security properties compared to the TMR, allowing optimized access patterns for untrusted applications while maintaining strict security for the TMR. This localized security zone enables efficient processing where needed while protecting sensitive areas.

Inventive Principle:
Principle #3Local quality

3Reliability

If fenced memory region is implemented to control access, then security is improved, but system complexity increases

Engineering Contradiction:
ImprovesecurityVSAvoidsystem complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The fenced memory region is designed as a universal security mechanism that can be used across multiple applications and data types. Rather than implementing separate security systems for different access control needs, the FMR provides a unified interface that handles various access scenarios through a single standardized mechanism, reducing overall system complexity while maintaining strong security.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP3899768B1Secure computer vision processing
Publication Date: 2025.10.15 ATI TECHNOLOGIES ULC
  • EP3899768B1 patent drawingFigure 1
  • EP3899768B1 patent drawingFigure 2
  • EP3899768B1 patent drawingFigure 3

AI summary

A computer vision processor (120) in an image cluster (105) defines a fenced memory region (FMR) that controls access to image data stored in a first portion of a trusted memory region (TMR)(125). The computer vision processor receives FMR requests from an application (150) implemented in a processing cluster (145). The FMR requests are to access the image data (130) in the first portion of the TMR. The computer vision processor selectively allows the requesting application to access the image data. In some cases, the computer vision processor acquires the image data and stores the image data in the first portion of the TMR, such as buffers in the TMR. A data fabric (135) selectively permits the image processing application to access the data stored in the TMR based on whether the image cluster has opened or closed the FMR for the portion of the TMR.