数字IC设计验证：Writing Testbenches第二版

需积分: 12 194 浏览量更新于2024-07-22 1 收藏 5.51MB PDF 举报

"Writing Testbenches 2nd" 是一本关于数字集成电路设计验证的经典书籍，作者Janick Bergeron在Qualis Design Corporation工作。这本书是第二版，专注于Verilog语言，同时也有VHDL和SystemVerilog版本供不同需求的读者选择。在数字集成电路设计中，测试平台（Testbench）的编写是验证硬件描述语言（HDL，如Verilog）模型功能是否正确的重要环节。这本书详细介绍了如何构建有效的测试平台来进行功能验证，确保设计的IC在实际应用中能够按照预期工作。测试平台通常用于模拟真实环境，通过提供输入信号并捕获输出结果，来评估设计的功能性和性能。第一版到第二版的更新可能包括了新的验证方法、技术进步以及对Verilog语言的深入解析。Janick Bergeron作为作者，可能分享了他在验证领域的专业经验和最佳实践，帮助读者提高验证效率，避免常见的设计错误。此书涵盖了多个关键知识点，例如： 1. **验证基础**：介绍验证的基本概念和流程，包括为何验证至关重要，以及如何规划和组织验证工作。 2. **Verilog测试平台设计**：讲解如何使用Verilog构建模块化的、可重用的测试平台，以适应不断变化的设计需求。 3. **激励生成**：探讨如何创建随机或确定性的激励序列，以充分覆盖设计的各种行为状态。 4. **断言与覆盖率**：解释如何利用断言来定义期望的行为，并衡量验证的完整性。 5. **环境构造**：介绍如何构造复杂的验证环境，包括激励混合器、代理、监控器和DUT接口等组件。 6. **高级验证技术**：可能包括基于UVM（Universal Verification Methodology）的验证方法学，以及使用OOP（面向对象编程）原则进行验证设计。 7. **调试与问题解决**：分享如何有效地调试测试平台和设计问题，以及如何记录和报告验证结果。 8. **案例研究**：通过具体的例子和项目，帮助读者理解和应用所学的知识。这本书对于从事IC设计和验证的专业人士，尤其是初学者来说，是一份宝贵的资源。它不仅提供了理论知识，还强调了实践经验，有助于提升读者在数字集成电路验证领域的技能。

What is Verification?

model of the universe as far as the design is concerned. The verifi-

cation challenge is to determine what input patterns to supply to the

design and what is the expected output of a properly- working

design when submitted to those input patterns.

Testbench

Design

under

Verification

Design

under

Verification

THE IMPORTANCE OF VERIFICATION

If you look at a typical book on Verilog or VHDL, you will find that

most of the chapters are devoted to describing the syntax and

semantics of the language. You will also invariably find two or

three chapters on synthesizeable coding style or Register Transfer

Level (RTL) subset.

Most often, only a single chapter is dedicated to testbenches. Very

little can be adequately explained in one chapter and these explana-

tions are usually very simplistic. In nearly all cases, these books

limit the techniques described to applying simple sequences of vec-

tors in a synchronous fashion. The output is then verified using a

waveform viewing tool. Most books also take advantage of the

topic to introduce the file input mechanisms offered by the lan-

guage, devoting yet more content to detailed syntax and semantics.

Given the significant proportion of literature devoted to writing

synthesizeable VHDL or Verilog code compared to writing test-

benches to verify their functional correctness, you could be tempted

to conclude that the former is a more daunting task than the latter.

The evidence found in all hardware design teams points to the con-

trary.

Today, in the era of multi-million gate ASICs, reusable intellectual

property (IP), and system-on-a-chip (SoC) designs, verification

consumes about 70% of the design effort. Design teams, properly

staffed to address the verification challenge, include engineers ded-

icated to verification. The number of verification engineers can be

up to twice the number of RTL designers.

2 Writing Testbenches: Functional Verification of HDL Models

Figure 1-1.

Generic

structure of a

testbench and

design under

verification

Most books focus

on syntax, seman-

tic and RTL sub-

set.

70% of design

effort goes to veri-

fication.

The Importance of Verification

Verification is on

the critical path.

Verification time

can be reduced

through parallel-

Verification time

can be reduced

through abstrac-

tion.

Using abstraction

reduces control

over low-level

details.

Verification time

can be reduced

through automa-

tion.

Given the amount of effort demanded by verification, the shortage

of qualified hardware design and verification engineers, and the

quantity of code that must be produced, it is no surprise that, in all

projects, verification rests squarely in the critical path. The fact that

verification is often considered after the design has been completed,

when the schedule has already been ruined, compounds the prob-

lem. It is also the reason verification is currently the target of new

tools and methodologies. These tools and methodologies attempt to

reduce the overall verification time by enabling parallelism of

effort, higher abstraction levels and automation.

If efforts can be parallelized, additional resources can be applied

effectively to reduce the total verification time. For example, dig-

ging a hole in the ground can be parallelized by providing more

workers armed with shovels. To parallelize the verification effort, it

is necessary to be able to write—and debug—testbenches in paral-

lel with each other as well as in parallel with the implementation of

the design.

Providing higher abstraction levels enables you to work more effi-

ciently without worrying about low-level details. Using a backhoe

to dig the same hole mentioned above is an example of using a

higher abstraction level.

Higher abstraction levels are usually accompanied by a reduction in

control and therefore must be chosen wisely. These higher abstrac-

tion levels often require additional training to understand the

abstraction mechanism and how the desired effect can be produced.

Using a backhoe to dig a hole suffers from the same loss-of-control

problem: The worker is no longer directly interacting with the dirt;

instead the worker is manipulating levers and pedals. Digging hap-

pens much faster, but with lower precision and only by a trained

operator. The verification process can use higher abstraction levels

by working at the transaction- or bus-cycle levels (or even higher

ones), instead of always dealing with low-level zeroes and ones.

Automation lets you do something else while a machine completes

a task autonomously, faster and with predictable results. Automa-

tion requires standard processes with well-defined inputs and out-

puts. Not all processes can be automated. For example, holes must

be dug in a variety of shapes, sizes, depths, locations and in varying

Writing Testbenches: Functional Verification of HDL Models

What is Verification?

soil conditions, which render general-purpose automation impossi-

ble.

Randomization

can be used as an

automation tool.

Verification faces similar challenges. Because of the variety of

functions, interfaces, protocols and transformations that must be

verified, it is not possible to provide a general purpose automation

solution for verification, given today's technology. It is possible to

automate some portion of the verification process, especially when

applied to a narrow application domain. For example, trenchers

have automated digging holes used to lay down conduits or cables

at shallow depths. Tools automating various portions of the verifi-

cation process are being introduced. For example, there are tools

that will automatically generate bus-functional models from a

higher-level abstract specification.

For specific domains, automation can be emulated using random-

ization. By constraining a random generator to produce valid inputs

within the bounds of a particular domain, it is possible to automati-

cally produce almost all of the interesting conditions. For example,

the tedious process of vacuuming the bottom of a pool can be auto-

mated using a broom head that, constrained by the vertical walls,

randomly moves along the bottom. After a few hours, only the cor-

ners and a few small spots remain to be cleaned manually. This type

of automation process takes more computation time to achieve the

same result, but it is completely autonomous, freeing valuable

resources to work on other critical tasks. Furthermore, this process

can be parallelized

easily by concurrently running several random

generators. They can also operate overnight, increasing the total

number of productive hours.

1. Optimizing these concurrent processes to reduce the amount of overlap

is another question!

4 Writing Testbenches: Functional Verification of HDL Models

Reconvergence Model

RECONVERGENCE MODEL

Do you know what

you are actually

verifying?

The reconvergence model is a conceptual representation of the veri-

fication process. It is used to illustrate what exactly is being veri-

fied.

One of the most important questions you must be able to answer is:

What are you verifying? The purpose of verification is to ensure

that the result of some transformation is as intended or as expected.

For example, the purpose of balancing a checkbook is to ensure that

all transactions have been recorded accurately and confirm that the

balance in the register reflects the amount of available funds.

Figure 1-2.

Reconvergent

paths in

verification

Transformation

Verification

Verification is the

reconciliation,

through different

means, of a speci-

fication and an

output.

Figure 1-2 shows that verification of a transformation can be

accomplished only through a second reconvergent path with a com-

mon source. The transformation can be any process that takes an

input and produces an output. RTL coding from a specification,

insertion of a scan chain, synthesizing RTL code into a gate-level

netlist and layout of a gate-level netlist are some of the transforma-

tions performed in a hardware design project. The verification pro-

cess reconciles the result with the starting point. If there is no

starting point common to the transformation and the verification, no

verification takes place.

The reconvergent model can be described using the checkbook

example as illustrated in Figure 1-3. The common origin is the pre-

vious month's balance in the checking account. The transformation

is the writing, recording and debiting of several checks during a

Writing Testbenches: Functional Verification of HDL Models

one-month period. The verification reconciles the final balance in

the checkbook register using this month's bank statement.

Figure 1-3.

Balancing a

checkbook is a

verification

process

Balance from

last month's

statement

Recording Checks

Reconciliation

Balance from

latest

statement

THE HUMAN FACTOR

If the transformation process is not completely automated from end

to end, it is necessary for an individual (or group of individuals) to

interpret a specification of the desired outcome and then perform

the transformation. RTL coding is an example of this situation. A

design team interprets a written specification document and pro-

duces what they believe to be functionally correct synthesizeable

HDL code. Usually, each engineer is left to verify that the code

written is indeed functionally correct.

Figure 1-4.

Reconvergent

paths in

ambiguous

situation

Specifi-^ Interpre-

cation ^ tation

RTL coding

Verifying your

own design veri-

fies against your

interpretation, not

against the specifi-

cation.

Figure 1-4 shows the reconvergent path model of the situation

described above. If the same individual performs the verification of

the RTL coding that initially required interpretation of a specifica-

tion, then the common origin is that interpretation, not the specifi-

cation.

In this situation, the verification effort verifies whether the design

accurately represents the implementer's interpretation of that speci-

fication. If that interpretation is wrong in any way, then this verifi-

cation activity will never highlight it.

Any human intervention in a process is a source of uncertainty and

unrepeatability. The probability of human-introduced errors in a

Writing Testbenches: Functional Verification of HDL Models

process can be reduced through several complementary mecha-

nisms: automation, poka-yoke or redundancy.

Automation

Eliminate human

intervention.

Automation is the obvious way to eliminate human-introduced

errors in a process. Automation takes human intervention com-

pletely out of the process. However, automation is not always pos-

sible, especially in processes that are not well-defined and continue

to require human ingenuity and creativity, such as hardware design.

Poka-Yoke

Make human inter-

vention foolproof.

Another possibility is to mistake-proof the human intervention by

reducing it to simple, and foolproof steps. Human intervention is

needed only to decide on the particular sequence or steps required

to obtain the desired results. This mechanism is also known as

poka-yoke in Total Quality Management circles. It is usually the

last step toward complete automation of a process. However, just

like automation, it requires a well-defined process with standard

transformation steps. The verification process remains an art that,

to this day, does not yield itself to well-defined steps.

Redundancy

Have two individ-

uals check each

other's work.

The final alternative to removing human errors is redundancy. It is

the simplest, but also the most costly mechanism. Redundancy

requires every transformation resource to be duplicated. Every

transformation accomplished by a human is either independently

verified by another individual, or two complete and separate trans-

formations are performed with each outcome compared to verify

that both produced the same or equivalent output. This mechanism

is used in high-reliability environments, such as airborne and space

systems. It is also used in industries where later redesign and

Writing Testbenches: Functional Verification of HDL Models 7

replacement of a defective product would be more costly than the

redundancy itself, such as ASIC design.

Figure 1-5.

Redundancy in

an ambiguous

situation

enables

accurate

verification

Interpre

tation

Specifi-,

cation

Interpre-

tation

RTL coding

A different person

should be in

charge of verifica-

tion.

Figure 1-5 shows the reconvergent paths model where redundancy

is used to guard against misinterpretation of an ambiguous specifi-

cation document. When used in the context of hardware design,

where the transformation process is writing RTL code from a writ-

ten specification document, this mechanism implies that a different

individual must be in charge of the verification.

WHAT IS BEING VERIFIED?

Choosing the common origin and reconvergence points determines

what is being verified. These origin and reconvergence points are

often determined by the tool used to perform the verification. It is

important to understand where these points lie to know which trans-

formation is being verified. Formal verification, model checking,

functional verification, and rule checkers verify different things

because they have different origin and reconvergence points.

Formal Verification

Formal verifica-

tion does not elim-

inate the need to

write testbenches.

Formal verification is often misunderstood initially. Engineers

unfamiliar with the formal verification process often imagine that it

is a tool that mathematically determines whether their design is cor-

rect, without having to write testbenches, Once you understand

what the. end points of the formal verification reconvergent paths

are, you know what exactly is being verified.

The application of formal verification falls under two broad catego-

ries: equivalence checking and model checking.

8 Writing Testbenches: Functional Verification of HDL Models

What Is Being Verified?

Equivalence Checking

Equivalence

checking com-

pares two models.

Figure 1-6 shows the reconvergent path model for equivalence

checking. This formal verification process mathematically proves

that the origin and output are logically equivalent and that the trans-

formation preserved its functionality.

Figure 1-6.

Equivalence

checking paths RTL or

Netlist

Synthesis

Equivalence

Checking

RTL or

Netlist

It can compare two

netlists.

It can detect bugs

in the synthesis

software.

In its most common use, equivalence checking compares two

netlists to ensure that some netlist post-processing, such as scan-

chain insertion, clock-tree synthesis or manual modification

, did

not change the functionality of the circuit.

Another popular use of equivalence checking is to verify that the

netlist correctly implements the original RTL code. If one trusted

the synthesis tool completely, this verification would not be neces-

sary. However, synthesis tools are large software systems that

depend on the correctness of algorithms and library information.

History has shown that such systems are prone to error. Equiva-

lence checking is used to keep the synthesis tool honest. In some

rare instances, this form of equivalence checking is used to verify

that manually written RTL code faithfully represents a legacy gate-

level design.

Less frequently, equivalence checking is used to verify that two

RTL descriptions are logically identical, sometimes to avoid run-

ning lengthy. regression simulations when only minor non-func-

tional changes are made to die source code to obtain better

synthesis results, or when a design is translated from an HDL to

another.

2. Text editors remain the greatest design tools!

Writing Testbenches: Functional Verification of HDL Models 9

vviiaiis v eruieaiion.'

Equivalence

checking found a

bug in an arith-

metic operator.

Equivalence checking is a true alternative path to the logic synthe-

sis transformation being verified. It is only interested in comparing

Boolean and sequential logic functions, not mapping these func-

tions to a specific technology while meeting stringent design con-

straints. Engineers using equivalence checking found a design at

Digital Equipment Corporation (now part of HP) to be synthesized

incorrectly. The design used a synthetic operator that was function-

ally incorrect when handling more than 48 bits. To the synthesis

tool's defense, the documentation of the operator clearly stated that

correctness was not guaranteed above 48 bits. Since the synthesis

tool had no knowledge of documentation, it could not know it was

generating invalid logic. Equivalence checking quickly identified a

problem that could have been very difficult to detect using gate-

level simulation.

Model Checking

Model checking

proves assertions

about the behavior

of the design.

Model checking is a more recent application of formal verification

technology. In it, assertions or characteristics of a design are for-

mally proven or disproved. For example, all state machines in a

design could be checked for unreachable or isolated states. A more

powerful model checker may be able to determine if deadlock con-

ditions can occur.

Another type of assertion that can be formally verified relates to

interfaces. Using a formal description language, assertions about

the interfaces of a design are stated and the tool attempts to prove or

disprove them. For example, an assertion might state that, given

that signal ALE will be asserted, then either the DTACK or

ABORT signal will be asserted eventually.

Figure 1-7.

Model

checking paths

RTL Coding

Specifi-

cation

.Interpretation

Model

Checking

Assertions

RTL

10 Writing Testbenches: Functional Verification of HDL Models

What Is Being Verified?

Knowing which

assertions to prove

and expressing

them correctly is

the most difficult

part.

The reconvergent path model for model checking is shown in

Figure 1-7 . The greatest obstacle for model checking technology is

identifying, through interpretation of the design specification,

which assertions to prove. Of those assertions, only a subset can be

proven feasibly. Current technology cannot prove high-level asser-

tions about a design to ensure that complex functionality is cor-

rectly implemented. It would be nice to be able to assert that, given

specific register settings, a set of asynchronous transfer mode

(ATM) cells will end up at a set of outputs in some relative order.

Unfortunately, model checking technology is not at that level yet.

Functional Verification

Figure 1-8.

Functional

verification

paths

Specifi-

cation

RTL Coding

RTL

Verification

Functional verifi-

cation verifies

design intent.

You can prove the

presence of bugs,

but you cannot

prove their

absence.

The main purpose of functional verification is to ensure that a

design implements intended functionality. As shown by the recon-

vergent path model in Figure 1-8, functional verification reconciles

a design with its specification. Without functional verification, one

must trust that the transformation of a specification document into

RTL code was performed correctly, without misinterpretation of the

specification's intent.

It is important to note that, unless a specification is written in a for-

mal language with precise semantics, it is impossible to prove that

a design meets the intent of its specification. Specification docu-

ments are written using natural languages by individuals with vary-

ing degrees of ability in communicating their intentions. Any

document is open to interpretation. Functional verification, as a

process, can show that a design meets the intent of its specification.

But it cannot prove it. One can easily prove that the design does not

implement the intended function by identifying a single discrep-

3. Even if such a language existed, one would eventually have to show

that this description is indeed an accurate description of the design

intent, based on some higher-level ambiguous specification.

Writing Testbenches: Functional Verification of HDL Models 11

剩余254页未读，继续阅读

fanlianglove

粉丝: 1
资源: 1

数字IC设计验证：Writing Testbenches第二版

writing_testbenches_2rd

writing testbench

Writing_Testbenches

履带式拖拉机Creo2.0_三维3D设计图纸.zip

SSM+JSP高校毕业生就业满意度调查统计系统答辩PPT.pptx

SSM+JSP冰淇淋在线购买网站答辩PPT.ppt

SSM+JSP医护系统答辩PPT.pptx

Shapely-1.6.4.post2-cp35-cp35m-win_amd64.whl

SSM+JSP农场信息化管理系统答辩PPT.pptx

基于谱聚类滤波器级剪枝方法用于压缩卷积神经网络

最新资源