Data Fabric Lightweight C-State Power Management
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In complex system-on-chip (SOC) designs, implementing power-saving modes is challenging due to the diverse and often isochronous operations of peripheral controllers, which hinder the achievement of chip-wide low-power states, especially when idle periods are short and exit latencies from low-power states are too long to accommodate real-time peripherals.
Innovation Solution
The introduction of a lightweight C-state for the data fabric with reduced exit latency, along with techniques like disconnect monitoring and C-state limit, allows the data fabric to autonomously enter a low-power state without intervention from the system management unit, and prevents entry into traditional C-states when CPU cores and GPU enter deep idle states, thereby reducing power consumption while maintaining low latency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If the data fabric enters a traditional low-power C-state to reduce power consumption, then power savings are achieved, but the exit latency becomes too long to accommodate real-time peripheral operations
Solution Approach 1:
The patent segments the traditional C-state into two distinct low-power states: a light-weight C-state with fast exit latency for real-time operations and a deep C-state with slow exit latency for non-real-time operations. This segmentation allows the data fabric to select the appropriate state based on the operational requirements of peripheral controllers, thereby resolving the contradiction between power savings and exit latency.
Solution Approach 2:
The system dynamically transitions between different C-states based on real-time traffic patterns and operational requirements. The data fabric can switch between the light-weight C-state and deep C-state depending on whether real-time peripheral operations are expected, making the power management adaptive rather than static. This dynamic approach allows optimization of both power consumption and exit latency based on actual system needs.
2Speed
If the data fabric remains active to quickly service requests from peripheral controllers, then real-time operations are supported, but power consumption increases
Solution Approach 1:
The data fabric employs dynamic power management by transitioning between active state, light-weight C-state, and deep C-state based on detected traffic patterns. When idle periods are detected, the system can enter the light-weight C-state which maintains fast response capability for real-time operations while consuming significantly less power than the active state. This dynamic state selection resolves the contradiction between maintaining fast response speed and reducing power consumption.
Solution Approach 2:
The patent changes the operational parameters of the data fabric by introducing a light-weight C-state with specific characteristics (fast exit latency, moderate power savings) that differ from both the active state and the traditional deep C-state. This parameter change enables the system to operate in a middle ground that balances response speed and power consumption, allowing the data fabric to service real-time requests efficiently while consuming less power than when fully active.
3Stability of the object's composition
If peripheral controllers use large buffers to accommodate long exit latencies, then data transfer continuity is maintained, but buffer size and memory requirements increase
Solution Approach 1:
The patent segments the power states to include a light-weight C-state with fast exit latency that is suitable for real-time peripheral operations. This segmentation allows peripheral controllers to use smaller buffers because the data fabric can quickly exit from the light-weight C-state when data is needed, maintaining data transfer continuity without requiring large buffer sizes to compensate for long exit latencies.
Data Source
AI summary
A data processor includes a plurality of requestors, a plurality of responders, and a data fabric. The data fabric is for routing requests between the plurality of requestors and the plurality of responders and has a plurality of non-operational power states including a normal C-state and a light-weight C-state. The light-weight C-state has lower entry and exit latencies than the normal C-state. The data fabric monitors traffic through the data fabric and places the data fabric in the light-weight C-state in response to detecting an idle traffic state.