Design Article

Opportunities in Analog Verification

Saranyan Vigraham, Qualcomm

10/28/2008 9:19 AM EDT

The wireless industry is continuously innovating and re-shaping the state-of-the-art techniques in analog and RF circuit design. The analog systems are getting increasingly challenging to design and even more to verify. The industry is seeing a bigger risk of functional failure in these systems or sub-systems as compared to the past decade. Functional failures can mean any of the following things and more -- Chip not powering on, inverted logic due to improper connectivity, clocks not propagating all the way to a core block, registers not getting updated, wrong modes of operation, incorrect performance in some of the gain modes in an amplifier, etc.

Normally speaking, these things should never happen and a chip must be verified for all these factors before tapeout. Usually, the circuit designers do this in the analog domain. However, these days an increasing demand for specialized verification engineers has been noted in the industry. The reasons primarily being the increasing complexity of block interactions that will cause "accidental" goof-up in connections or logic, and an alternate mindset and skills needed to do the job.

Verification engineers have to be capable of thinking beyond a particular block and formulate their verification strategies in the context of the complete system. Also, verification engineers must be able to abstract a system to the simplest form needed to conduct the tests. Programming skills, scripting and little bit of design skills go a great deal in forming a strong verification engineer. There are different aspects of system verification -- performance and functional.

Performance verification is to verify if the block matches all the numbers from an objective specification document. In case of RF systems this can be compression points, matching, distortion, gain, noise figures, etc. Analog systems might have specifications like gain and noise figure. Whatever these may be, the designers usually do an excellent job in making sure their block meets the required performance. They run their circuits on Process, Voltage and Temperature corners and use different transistor models to characterize the performance of their circuits to the death. However, most of them find it hard to think beyond their blocks. The best block in the system will not do any good if the chip fails to power up. Also, a block like Programmable Gain Amplifier (PGA) might have an eight bit digital control that sets the gain. It is important the effect of all the 256 values. Most of the times, the designers cannot get to this because it is not feasible to run 256 transistor level simulations for testing gain modes when the deadlines force to work on other "important" things.

The verification engineer mindset should be different and they should be told that their only job is to find faults or cases that break a design. As mentioned above, these are functional verification tests. Hardware Description Languages (HDL) like Verilog-AMS helps a great deal in this challenging task. Using the HDL, the most complex blocks can be abstracted to represent its functional behavior with few lines of code. Assertions (error checking conditions) can be included in these models. These assertions perform beautifully in flagging any error. For instance, assertions can used to pull the output of a block to ground if the block is not powered on or if the bias conditions are not in the proper ranges or even if the digital inputs seen by the block are not expected values. This will break the signal path's functionality and thereby make issues like connectivity, incorrect power supplies, and improper biasing standout. Unless a chip passes the final functional verification, it cannot be taped out.

Presently, specialized staffing for analog verification is hard to find. Hence, automating some of the aforementioned tasks and tool support will play a big part in how this process evolves. It is a normal belief in the industry that for every five design engineers, one modeling and verification engineer is required. However, this rarely happens. While this concept is just starting to get popular, verification engineers today typically support anywhere between 10-20 designers. This is a hard task. The EDA companies have already realized this opportunity and incorporating automatic HDL model generation capabilities in their tools (the latest release of Virtuoso from Cadence is an example). However, the effort is still akin to going after the low-hanging fruits than solving the bigger issues. I will present more on this issue later.

One issue that presents a formidable challenge to analog verification is the lack of a framework for top down design. Top down design has been effectively used in digital design flow for over a decade. The quality of initial tapeouts in digital design is way superior to their analog counterparts.

A top down design flow for analog designs would take the following steps:

  • Define project guidelines or rules. These rules range from voltage domain information to pin naming conventions. The naming conventions, if followed correctly will make the verification easy after integration of all the blocks.
  • Form a hierarchy of the complete analog chip at least until the top level description (or schematics) of core blocks. For instance, in a wireless receiver system, this might mean deciding architecture, blocks and IPs to be used (both new and from existing work), determine the connectivity and floor plan. If it is a heterodyne receiver, what sort of LNA are we looking at, what should be the linearity of the mixer, etc.
  • If the chip uses digital control and has local registers, floor plan them and define the digital control modes using a Finite State Machine (later modeled using a HDL like Verilog).
  • Write behavioral models for all the core blocks. Using any form of HDL that is suitable (Verilog-AMS, Verilog-A, SystemVerilog, etc) to create these behavioral models.
  • Simulate the complete signal path. Once the chinks have been worked out and architecture formalized, start transistor level design for the core blocks.
With the tools and resources at our disposal, the above steps are hard to accomplish before starting detailed transistor level circuit design. The challenging project deadlines mean that the above tasks should take no longer than couple of weeks. Staffing becomes critical in the initial phase of the project. The system, modeling and verification engineers/leads work together with the chip lead to accomplish all the above tasks in about two weeks. After this has been accomplished, the transistor design will begin within a framework that has been setup for the project. In the next article, we will demonstrate how to achieve the aforementioned tasks with the support of some existing CAD tools, scripting and programming solutions, and more importantly common sense and foresight.

About the Author:
Saranyan Vigraham
is a Senior Engineer with RF Analog group at Qualcomm, Austin. He holds a master and a PhD in Computer Engineering from Wright State University.





Mike_Demler

10/28/2008 3:30 PM EDT

Hello Saranyan,

It is interesting to read your perspective on the challenges of RF/AMS verification. That is the main subject of my blog, so I invite you to check it out, and participate in the discussion there as well.

Regards,
Mike Demler
href='http://synopsysoc.org/analoginsights/'>Analog Insights

Sign in to Reply



Saranyan

10/28/2008 4:30 PM EDT

Mike, I will check it out. Thanks for the invite.

Sign in to Reply



eletste

11/19/2008 4:22 AM EST

Hello Saranyan,

I'm currently working as mixed-signal designer and I'm trying to bring within my team the same design/verification guidelines you mentioned in your article. In fact with the tools we have now available, it is impossible to fully verify a design with heavy analog and digital sections at transistor level. I'm glad that someone in pointing out the necessity to have a dedicated person for this purpose.

thanks,

Stefano Camera
STMicroelectronics

Sign in to Reply



pitchMonk1

11/19/2008 1:26 PM EST

Saranyan, Interesting article. Some of the sentiments are already echoed by Henry Change in Design Guide Consulting. While the suggested solution is seemingly easy, analog designers worry about a number of things to make sure that the chip/IP works. Many of the passing metrics are "qualifiers" which cannot be brought under a pass/fail criteria. This is the same issue with debug as well. I think we are in an inflection point where designers are relooking at the way they do analog. Two changes have happened to prompt this:
(1) Chips are becoming more and more mixed-signal.
(2) Analog designers cannot design anymore in isolation. They have to constantly interact with digital designers and use digital transistors for their design.

Thanks Mike for the invitation to your Synopsys board. I am sure you write something exciting Synopsys is doing.

Sign in to Reply



sunnylvee

12/1/2008 4:42 PM EST

Saranyan, it is really nice to read such a good article. PitchMonk1, thanks for mentioning Designer's Guide Consulting, yes, we are doing analog verification now and we have helped customers find bugs either in analog blocks or digital blocks before the tape-out. if you want to know more information about this. just click http://www.designers-guide.com/

Thanks
Sunny Zhang
Designer's Guide Consulting

Sign in to Reply



Sonali Luniya

12/3/2008 10:19 AM EST

Hello Saranyan,
This is an interesting article. There is one question I would like to ask is who is the perfect candidate for this job? I worked at RFMD (for 10 months) until early this year when they went through a round of lay offs. I was the CAD and modeling engineer for the RF team. I developed the AMS models for the all the RF and Front End Module blocks and performed functional verification mixed signal simulation of the entire signal path. I did not have an opportunity to develop test benches and automate the tests since there was an entire verification team dedicated to do that.

I have been looking for a job since June. I did manage to get an offer from a cell phone giant but the position did not get approved since they merged with another semiconductor giants. The other companies I interviewed with came back saying we need somebody with more than just VerilogAMS and simulation expertise, somebody with more verification and scripting experience. From my experience at RFMD one person cannot do verification, scripting, simulation, modeling, debug test cases, understand every design block and simulation tool/technique.
Are there jobs out there for a beginner verification engineer with analog/RF modeling background and can grow into the verification engineers role?
Thanks,
Sonali Luniya.

Sign in to Reply



Babun

12/8/2008 12:12 AM EST

Hi pitchMonk1,

Thanks for your comments. Can you please explain in details the point 2 in your comment.

Thank you.

Best Regards,
Babun

Sign in to Reply



sunnylvee

12/11/2008 3:16 PM EST

Hi Sonali,

You can get all the answers about analog verification from the training classes held by Designers' Guide Consulting. There will be one in next January. Just click http://www.designers-guide.com/classes.html
and you will know more.

Best Wishes

Sunny

Sign in to Reply



Saranyan

12/12/2008 12:30 AM EST

Everyone - sorry for the delay in responding.

@Stefano - Yes, you are right. It is challenging to verify the current designs due to complexity. The one big problem is that the present EDA tools do not focus on methodologies. Actually, it is not about the EDA companies, analog designers and chip leads have to bring a methodical approach to verifying their chips. Simple things like naming conventions, an effort to top down design and modular and early verification plans would help significantly.

@Sonali - Industry can be brutal. I do not fully agree with your claim that one person cannot do everything (verification, scripting, etc). Actually, the way I see them, these skills are actually weapons in a verification engineers arsenal to get their jobs done smoothly and efficiently.
Regarding jobs purely for RF/Analog modeling engineers, I don't know, because modeling is a small part of verification. Verification engineers need to model blocks with various levels of abstraction, plan test benches and anticipate where the chip might be functionally unstable and test for those cases. There are several opportunities available for engineers with a knowledge in RF/Analog modeling and a willingness to learn more.

@Babun - which point of Stefano's reply do you want elaborated?

-Saranyan

Sign in to Reply



Saranyan

12/13/2008 11:45 AM EST

I am sorry.
@Babun - I think what pitchmonk means is that analog designers have to think beyond their blocks. They have to think about interfaces to their block and the control signal they get. I am not sure about what he means by digital transistors though.

Sign in to Reply



Please sign in to post comment

Navigate to related information

Datasheets.com Parts Search

185 million searchable parts
(please enter a part number or hit search to begin)

Feedback Form