How Memory Devices Dynamically Manage Overly Busy Data Rows

This patent describes a method for memory devices to identify and manage frequently accessed data rows by dynamically adjusting a threshold based on how many such 'aggressor rows' are already detected.

What it covers

A memory device detects 'aggressor rows' by tracking how often individual memory rows are accessed. When a row is accessed, a count value associated with that row changes. If this count value exceeds a 'mitigation threshold', the row is identified as an aggressor. These identified aggressor rows are then stored in an 'aggressor queue'. The system intelligently adjusts the 'mitigation threshold' based on the current number of addresses already stored in the aggressor queue. For example, the abstract states that the threshold may increase as more addresses are added to the queue, making it harder for new rows to be flagged as aggressors when many are already identified.

What it doesn't cover

• —Does not cover memory management techniques that do not involve tracking individual row access counts.
• —Does not cover systems that use a fixed or static mitigation threshold for identifying aggressor rows.
• —Does not cover methods for mitigating memory issues that do not involve an 'aggressor queue' to store identified rows.
• —Does not cover identifying aggressor rows based on factors other than a changing count value compared to a threshold.

The clever bit

The truly novel aspect is the dynamic adjustment of the 'mitigation threshold' based on the number of aggressor rows already detected and queued. This allows the system to adapt its sensitivity, preventing over-mitigation when only a few rows are busy, and becoming more selective when many rows are already flagged.

Why it matters

Modern memory devices, especially DRAM, can suffer from reliability and performance issues when certain data rows are accessed too frequently, a phenomenon sometimes called 'Rowhammer'. This patent offers a way to intelligently detect and manage these 'aggressor rows' to prevent performance degradation or potential data corruption. By dynamically adjusting the detection sensitivity, it aims to optimize memory performance and longevity in high-demand computing environments.

Real-world examples

• 1.DRAM modules in servers and personal computers
• 2.Memory controllers in high-performance computing systems
• 3.Embedded memory in mobile devices
• 4.Solid-state drive (SSD) controllers