ASIC design is a complex, multistage, time-consuming
process, with huge associated development costs. It's important to get the
design right at the first attempt and also meet time-to-market criteria.
Verification demonstrates the functional correctness of a design but it consumes
almost 60 to 70% of the total resources in a typical chip design cycle. With
silicon design limits increasing 2x every 18 months, it is important to verify
functionality throughout the design to minimize the escalating cost of making a
mistake. One error can decide success or failure of a product.
An effective verification environment (VE) is just as essential as a basic
design environment. An effective VE can aid in the development of a single chip
and is imperative for creating multiple systems-on-chip (SoCs). Its value in the
future will only increase as chips become larger, more complex, and more
difficult to verify. Verification elements should be readily available in a
predefined, easy-to-use form. Like any IP, they must be modular, scalable, and
reusable. Present day SoC sizes run in hundreds of millions of gates, for
instance, Pentium IV comprises 55 million gates and network and backbone
processors, in excess of 100 million gates. They have become more sophisticated
and complex to verify. Today's SoCs can have integrated processor cores,
memories (RAMs, Cache, register banks), pipelines and parallel processing
blocks, customized modules and multiple interfaces.
Since the design complexity has drastically increased, verification at
different stages of ASIC development has become absolutely essential. Various
techniques and tools have evolved to support verification. There are different
types of verification depending on the development stage:
- Functional verification
- Assertion-based verification
- Formal verification
- Gate-level verification
- Wafer testing
- Chip bring-up verification
- Manufacturing verification
|
This paper will discuss functional, assertion-based and formal verification
types that deal with RTL and netlist verification.
Functional verification
Functional verification determines correctness of logic in design before it
is implemented on a device. It is used to check design compliance with
specifications. Functional verification takes place during the early stages of
the design flow so as to detect design flaws early in the chip development
process. With the ever increasing complexities and an increasing use of reusable
IP blocks, each new system IC and IP core has to be verified to eliminate
discrepancies between the design and its specifications to ensure that the
manufactured device functions properly. The test environment is developed and
designed as per specifications. It generates and drives the defined stimulus to
DUT (Device Under Test). Because timing information for the implemented design
is not available, simulation tests the logic in the design using unit delays.
Following are methods of carrying out functional verification.
Conventional Verification
Conventional verification is a BFM-based (bus functional model) approach where test benches are HDL
(Verilog/VHDL) or 'C'-based. The verification engineer using this approach
targets one bug/scenario at a time. Here, verification is restricted by the
verification approach and the verification engineer's visibility of the test
scenarios. But, attaining mere code coverage is no longer sufficient. Directed
test cases have lost relevance now that designs have become very complex and
SoCs have millions of gates. They are written only to fill remnants. And there
aren't enough hits with a single test run. Thousands of iterations are required
for full functional and code coverage. The major drawbacks of HDL-based
verification are:
- Test cases, as opposed to coveraged, are defined.
- Absence of random environment and random test case generation required to
fill maximum coverage.
|
Hardware verification language (HVL) based verification
Randomization, coverage, and analysis-based environments are implemented using
HVL-based verification. It decreases the time and effort to verify and increases
the verification strength. Only coverage groups need to be defined and the
compliance checklist mapped in them. Using HVLs, the verification engineer is
able to:
- Implement checkers, sequences.
- Generate signals, temporals.
- Inject and generate error/invalid scenarios and illegal conditions.
|
HVLs support use of coverage-driven verification methodology that enables
complete verification of DUT using automatically generated real world scenarios.
They also help create an integrated verification environment that supports
constrained random traffic generation and functional coverage. This increases
test range significantly and enhances maximum possibility for all kinds of
corner cases. Automatic generation of transactions implies automatic coverage of
corner cases and real world scenarios.
Benefits of HVL-based verification:
- Automatic, Random and Constrained Random generation
- Automatic generation of transactions possible.
- Automatic coverage of corner case scenarios and unknown corner cases.
- Random packet generation and transmission resulting in greater possibility
of hitting unexpected cases.
- Random and constrained random generation supported.
- Ranges, rules on generation possible for directed random packet generation.
- Unknown corner cases covered .
- Coverage Driven Verification
- Automated random generation functionality coverage.
- Measure of verification completeness provided by functional coverage.
- Corner cases exercised, with addition of a few directed test cases.
|
|
|
Figure 1. Constrained random generation improves coverage productivity |
VIP-based verification
Most designs today incorporate standard protocols. For example, PCI
Express, GBE, and USB are industry standards used in many chips. These standards
have to be implemented correctly for the success of the chip. Testing designs
against such verification IP to verify these standards increases verification
productivity considerably. IP/design reuse provides only 30 to 40% effort savings.
The big ROI lies in verification reuse. Verification consumes up to 70%
of the time and effort and if users are focused on design IP, they address only
30% of the issue. Verification IP offers remarkable value to teams trying
to integrate an IP core into their device. Until recently, VIP often only
consisted of a bus-functional model (BFM), and possibly included a monitor for
protocol checking. For VIP to be effective in enabling reuse of complex design
IP, it needs to provide many more features and functions. The main advantages of
using verification IP for verification are reduced time-to-verify, reduced
costs, a scalable verification environment that enables bottom-up verification
of complex SoCs, and fast ramp-up to functional testing stage.
Layered, object-oriented verification methodology for next generation
SoCs
Layer-based, object oriented verification environments are required to verify
today's complex chips/SoCs. Layered verification environments (VEs) are the key
to handling design modifications. VEs implemented layerwise, are modular,
scalable, and reusable for different design configurations. Environments
developed using languages that support layered, object oriented verification
(LOOV) provide scalability, modularity, and reusability. LOOV makes each layer
and module in the VE interoperable and interdependent. Information sharing
between modules and between layers becomes possible.
The elements of a verification environment are:
- Layers
- Signaling layer (Interface/BFM)
- Operational layer (Data manipulation / Data processing)
- Modules
- Signaling modules (Interface/BFM)
- Operational modules (Data manipulation / Data processing)
- Data modules
- Control module
- Test paths
- Test cases
| |
Reusable VEs serve both module and system-level verification. Scalable VEs
immediately acknowledge changes in functional modes or interfaces. For
layer-wise VE implementation, design changes require changing only the
corresponding VE layer. For example, if a new payload type is added, only the
Data Protocol layer must be modified.
|
|
Figure 2. Object-oriented verification-based verification
environment |
Assertion-based verification
Assertion-based verification is a methodology that allows assertions to be
specified. Assertions are design checks, either embedded deep in the design by a
designer to catch the bugs early on or some pin-checks written by a verification
engineer for black box testing of the code. These assertions detect bugs and
implement coverage that guide test case writers to develop relevant and more
effective tests. Their basic function is:
- Monitor all the interfaces to DUT in order to validate correctness of
stimulus.
- Write protocol monitors in the DUT so that it is continuously monitored
internally for the correct behavior.
- Write coverage properties to make sure current test suite covers all
required stimulus scenarios needed to test DUT functionality.
|
Assertions provide information about the functional/behavioral nature of the
block as the designer intended it to be used. They can also be thought of as
internal software test points that wait for a particular predefined condition to
occur and then notify the designer about the occurrence. This is much better
since the bugs are reported at the cause point and not the effect point, saving
the tedious process of tracing waveforms back to their original causes. Without
assertions, test vectors have to be much longer to ensure that triggered bugs are
propagated to observable outputs, else the errors remained undetected. By adding
assertions, behavior inside the design could be checked and bugs observed
instantly, at their source.
Assertion-based verification benefits users by:
- Simplifying diagnosis and detecting bugs by localizing occurrence of a
suspected bug, thereby reducing simulation-debug time significantly.
- Self-checking code helps in design reuse.
- Interface assertions help find interface issues early on.
|
Formal verification
Formal verification is an algorithmic-based, mathematical approach to logic
verification that thoroughly establishes the functional properties of a
design. It's an alternative approach to simulation for functional verification.
It proves that the origin and output are logically equivalent and that the
functionality matches. The two major types of formal verification used are
equivalency checking and formal RTL verification. Once the design is
functionally verified, formal verification is used to compare and verify the
design at various levels of abstraction. As the device size and complexity
increase, functional simulation becomes very time-consuming for both test bench
writing and simulation. Formal verification has a short run time and provides
complete functional coverage. Formal verification is a proven technology/flow in
an ASIC design environment.
Equivalency checking
Equivalency checking is the most successful and
easiest to implement formal technique. Equivalency checkers prove that one
gate or RTL design representation is equivalent to another gate or RTL
representation. This technique is used to compare a design's synthesized
gate-level netlist. Formal equivalence checking compares RTL, gate, gate with
clock trees, gate with scan, and gate with engineering change orders to the
"golden" RTL (an RTL design that has been fully verified to meet the design's functional requirements). The logical models of design primitives are formally compared with their transistor implementations, and they, in turn, are compared with the final layout representations. The primary use of equivalency checkers is to establish that successive design iterations still adhere to functionality of the golden RTL .
Formal RTL verification or model checking
Formal RTL verification tools are used to verify the correctness of the RTL.
The verified RTL is called the "Golden RTL." Functional formal verification uses
formal techniques to prove that a condition (often called property or assertion)
can or cannot exist for a given design implementation. In other words, it's used
to prove (or disprove) the assertions or characteristics of the design.
Functional formal verification tools assure that each check is proven for the
entire input space. Also, it overcomes the incompleteness issue in simulation
methods.
Conclusion
However, there still is much more to verification than meets the eye. As
mentioned before, verification consumes almost 60 to 70% of the total resources in
a typical chip design cycle. Chip design flow requires verification at each
level of abstraction beginning from architecture to silicon prototyping. The
challenge for engineers here is to verify spec-adherence for multiple
abstractions. The verification process depends on various parameters such as
the HVL used, the methodology employed, and the complexity of SoCs in terms of
size, functionality and application.
There are a plethora of new tools,
methodologies and languages in the semiconductor industry today. Each user
segment has it's own preference. The combination of languages, tools, IPs and
methodologies has morphed traditional verification into a "Hybrid Model"
process. Since time-to-verify has become a major concern, moving forward,
verification engineers may now have to depend on the "Hybrid Verification
Model," that addresses all parameters of the verification process.
The future of
design verification does not belong to any particular tool, language or
methodology, but a combination of them all.
By Rohit Dubey. Marketing Manager, ASIC Solutions & Services, eInfochips.
Go to the eInfochips, Ltd. website to learn more. |