技术文档_奇捷科技

Post-Mask ECO的补丁设计上预备资源的应用策略

下载文档

Resource Strategy for Post-Mask (Metal-Only) Functional ECO

EasylogicECO White Paper

WP-PMR-01

Functional ECO (Engineering Change Order) is an incremental

design method used to revise existing ASIC designs. Once an

RTL code is revised to change the ASIC function, the ECO

process modifies a portion of the existing netlist to align with

the revised ASIC function while preserving the integrity of

most of the netlist.

Functional ECO requests have become increasingly common

in recent years due to the rising complexity of ASIC design

and shorter project cycles. When an RTL change is necessary,

re-spinning the entire design always causes project delays. A

successful ECO task offers a shortcut for incorporating the

required changes swiftly. In the late design stage

(layout-completed or already taped-out), a successful ECO

task can significantly reduce the project delays from months

to just a few days, while maintaining the same quality results.

This article examines the available spare resources associated

with a post-mask functional ECO task and explores the

additional resources employed by EasylogicECO.

Introduction to Post-Mask ECO

The functional ECO process is divided into two primary aspects:

1. Pre-layout changes, and

2. Post-layout changes.

The necessary revisions in a functional ECO

task depend on the stage of the design project.

Pre-layout changes involve modifications to the

design logic and related implementations of

specific design requirements, such as low-power

cells, scan chain netlists, P&R data. On the other

hand, post-layout changes are limited to the

metal layers and aim to re-connect available logic

resources (referred to as Post-Mask ECO or Metal

ECO). Figure 1 illustrates the types of data that

require fixing.

Figure 1: Types of data to be revised in a post-mask ECO task

In this paper, our focus will be on the patch implementation and the resource strategy in a post-mask ECO.

Patch Implementation Criteria in Post-Mask Functional ECO Task

An efficient patch implementation algorithm needs to consider three factors:

1. Timing delay of the revised netlist

2. Available spare resources for patch creation

3. Size of ECO patch

The timing delay of the revised netlist, derived from

analyzing the original netlist, represents the target

timing to achieve. The patch logic of the revised

function must fulfill the timing requirement using

the existing spare resources, which are predetermined

in terms of cell functions, delays, locations, and

quantity. Consequently, the number of available

resources ultimately determines the outcome of the

patch, including both the timing delay and the

patch size.

That is why we focus on resource strategy as the

foundation for the success of post-mask ECO.

Figure 2: Trade-off during patch implementation

Overview of Spare Resources for Post-Mask ECO

If a design is layout-complete or already taped-out (so called post-mask), success heavily relies on the spare resources available for the patch implementation. The resources used to implement the patch need to be physically located in the vicinity of the ECO net where the function needs to be revised.

Spare resources for a post-mask functional ECO include four types of logic cells:

1. Spare cells

2. Gate arrays

3. Filler cells

4. Standard cells that are disengaged from the original

function as a result of the ECO

Traditional Approaches Employed in Handling Spare Resources

Traditional approaches involve two types of spare resources:

1. Spare cells

2. Gate arrays

(1) Spare Cells

Spare cells refer to a set of pre-designed standard cells that are not initially used in the main design but are reserved for post-mask modifications. They offer flexibility and enable functional modifications without significant layout changes. The specific types of spare cells used depend on the characteristics of the design, with control logic and combinational logic often requiring different types of logic cells in a functional ECO situation.

There is a limitation on the number of spare cells that can be placed, requiring careful allocation and potentially resulting in suboptimal usage of available resources.

(2) Gate Arrays

Gate arrays involve predefined arrays of basic logic gates that can be configured to implement specific functions by applying metal-only connections. Gate arrays can be used for targeted changes without requiring extensive design iterations.

However, gate arrays may have higher area and power overhead compared to standard cells. Their usage requires careful planning and integration into the design. Due to their larger size, gate arrays are often placed further away from dense routing areas, which can introduce timing delay issues."

Spare cells and gate arrays have pros and cons in terms of timing delay, area usage, and power consumption. The choice depends on the specific design application, as different applications may prefer a specific combination of resources. This white paper focuses solely on demonstrating the fundamental concepts and does not delve into design strategies.

Additional Spare Resources Utilized by EasylogicECO

As the number of available resources is crucial for the ECO success, EasylogicECO deploys an innovative algorithm to increase the resources for patch implementation. EasylogicECO utilizes the following two types of additional resources:

1. Filler cells

2. Standard cells that are disengaged from the original netlist as part of the functional ECO

This advanced approach significantly increases the number of resources and enhances the likelihood of achieving success in post-mask ECO.

(1) Filler Cells

Filler cells are additional cells incorporated into the layout to enhance manufacturing yield. They possess a complete gate structure with assigned VDD/VSS pins, adhering to the layout rules of a standard cell. However, they lack logical functionality as their metal layers intentionally remain open.

Filler cells that establish contact connections with poly, diffusion, and metal 1 layers can be utilized to construct functional gates. Like configuring gate arrays, filler cells use metal connection layers to configure the cell into specific functions. This technology is well-suited for functional ECO involving a small patch without necessitating extensive design alterations.

(2) Disengaged Standard Cells

Functional ECO cases often involve adding new logic gates while simultaneously removing some original gates. The removed gates are typically located near the ECO point, making them the optimal choice for implementing the patch. The reuse of these disengaged standard cells from the original netlist introduces an intriguing concept that offers valuable benefits in terms of routing delay and design flexibility.

Let's consider the following example: an ECO task involves adding 30 gates to create a patch while disengaging 20 original gates. EasylogicECO examines the removed circuit, and if 10 out of those 20 disengaged gates can be reused, it reduces the number of required spare cells from 30 to 20. This approach can double the success rate in certain cases.

Selection Techniques to Best Utilize Available Spare Resources

EasylogicECO employs a patented algorithm to optimize timing delays for post-mask functional ECO jobs. Alongside utilizing additional spare resources, the patch implementation process incorporates specific resource selection techniques to ensure an optimal optimization result. These techniques, illustrated in Figure 3, include:

1. Physical-aware patch generation

The physical information of the design

and available spare resources, obtained

from LEF/DEF input and library files, is used

to generate a patch that satisfies the timing

delay requirements of the design.

2. Number of logic levels

To meet timing requirements, EasylogicECO

analyzes the original design and the provided

physical information to determine the suitable

range of logic levels. This analysis result is then

used in patch optimization.

Figure 3: Resource selection techniques for best patch result

3. Wire delay

As wire delays play a critical role, EasylogicECO estimates the routing delay by leveraging the physical information from the netlist and the user-provided spare resources.

2. Routing congestion

Routability is taken into consideration during cell selections, incorporating a congestion factor within the optimization algorithm. This factor assesses the number of connected gates in the surrounding area of the target spare instance.

3. Driving strength

Based on the fanout requirements and estimated wire lengths, cell selections for the post-mask patch will prioritize cells with appropriate driving strength.

Conclusion

By utilizing additional spare resources and considering critical timing factors in the physical optimization for post-mask functional ECO, EasylogicECO efficiently delivers patches that meet the required timing delays while minimizing tool execution time.

About the author

Kager Tsai

Vice President of Technical Support, Easy-Logic Technology

Since joining Easy-Logic Technology in 2021, Kager has been instrumental

in leading his team to offer top-notch technical support to users globally.

With 18 years of experience in the CAD field and has worked for companies

such as SiS, MStar, MTK, and Cadence, Kager is proficient in the ASIC

front-end tool flow, especially in formal verification and ECO-related

applications. Kager also participated in the development of many

advanced ECO tool features.