On-Chip Cache Row-Column Addressing for Small-Packet Storage

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing chip designs face inefficiencies in storage space utilization and excessive power consumption due to the inability to effectively handle packets with small packet lengths, leading to wasted memory space and increased area requirements.

Innovation Solution

An on-chip cache apparatus utilizing a cache module and memory module to store packets using row and column addresses, allowing multiple packets to be cached and stored simultaneously, with conflict handling mechanisms to optimize space utilization and reduce power consumption.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If the chip increases dominant frequency and stacking resources to improve performance, then performance indicators improve, but chip area consumption increases excessively

Engineering Contradiction:
Improveperformance indicatorsVSAvoidchip area
Core Design Contradiction:
ProductivityVSArea of stationary object

Solution Approach 1:

The patent implements dynamic resource allocation through a state machine that monitors packet arrival patterns and dynamically switches between different cache modes (single-packet mode, multi-packet mode, and bypass mode). This allows the system to adapt to varying traffic conditions and optimize the balance between performance and area usage, rather than using a fixed high-performance design that consumes excessive area.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system changes operational parameters based on packet characteristics. When small packets are detected, the cache depth and allocation strategy are adjusted to accommodate multiple packets per row. The state machine transitions between different operational states (STATE_0, STATE_1, STATE_2) that modify cache behavior, enabling the same hardware to efficiently handle different packet sizes and patterns without requiring excessive resources.

Inventive Principle:
Principle #35Parameter changes

2Adaptability or versatility

If the chip design allocates large storage space to handle small packets, then small packet processing capability improves, but power consumption increases excessively

Engineering Contradiction:
Improvesmall packet processing capabilityVSAvoidpower consumption
Core Design Contradiction:
Adaptability or versatilityVSUse of energy by stationary object

Solution Approach 1:

The patent employs a state machine that dynamically adjusts cache behavior based on packet arrival patterns. When small packets are detected, the system transitions to multi-packet cache mode which efficiently utilizes storage space. When traffic patterns change, the state machine can switch to bypass mode or single-packet mode, reducing power consumption by deactivating or minimizing cache operations. This dynamic adaptation eliminates the need for continuously operating large cache structures.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent extracts and handles small packets differently from regular packets by implementing a specialized multi-packet mode. Instead of using the same cache structure for all packets, the system identifies small packets and routes them through a optimized path that consolidates multiple packets into fewer cache rows. This extraction of the small packet handling function reduces the overall cache size needed and lowers power consumption.

Inventive Principle:
Principle #2Taking out (Extraction)

3Device complexity

If each row stores only one packet according to row address, then address management simplifies, but storage space utilization deteriorates

Engineering Contradiction:
Improveaddress management complexityVSAvoidstorage space utilization
Core Design Contradiction:
Device complexityVSQuantity of substance

Solution Approach 1:

The patent segments the cache addressing into two levels: row address for selecting cache lines and column address for selecting specific packets within those lines. This segmentation allows multiple packets to be stored in a single row by using different column addresses. The state machine manages the complexity by automatically handling column address assignment when in multi-packet mode, maintaining relatively simple address management while dramatically improving storage utilization.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent adds a column address dimension to the traditional row-address-only cache structure. By introducing column addresses, the system can store multiple packets within a single row, effectively transforming from a one-dimensional (row-only) addressing scheme to a two-dimensional (row-column) scheme. This dimensional change enables efficient packing of small packets without significantly complicating the address management logic, as the state machine handles the coordination.

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

Data Source

PatentEP4137956B1On-chip cache device, on-chip cache read/write method, and computer readable medium
Publication Date: 2025.07.30 SANECHIPS TECH CO LTD
  • EP4137956B1 patent drawingFigure 1
  • EP4137956B1 patent drawingFigure 2
  • EP4137956B1 patent drawingFigure 3

AI summary

The present application provides an on-chip cache apparatus, an on-chip cache on-chip cache read-write method and a computer-readable medium, the on-chip cache apparatus includes: a read-write processing module, a cache module and a memory module; the read-write processing module is connected with the cache module and the memory module respectively, and is configured to store packets into the cache module and the memory module, read packets stored in the cache module and the memory module, and transfer packets cached in the cache module to the memory module for storing; the cache module is connected with the memory module through the read-write processing module, and includes at least one cache register configured to temporarily cache packets; and the memory module is connected with the read-write processing module, and is configured to store the packets cached in the cache module.