My Personal Journey

The Biggest Little Industry You’ve Never Heard of Turns Fifty

Posted on the HuffingtonPost earlier

Chips touch every aspect of our lives. You use chips in your car, in
your phone, in your TV, in your fridge, when you play a video game, when
you text, Skype or blog, in the bar code reader at the grocery
checkout, when you take a photo, as your luggage is routed through an
airport — any time you use electronics today you are using chips.

Now I am not talking about potato chips, I’m talking about
semiconductors — integrated circuits. Those small, intricate pieces of
silicon, doped with chemicals in factories in the U.S. and Taiwan, that
use logic and memory to take action for you. To shoot the zombie, or
control the brakes on your car. To route your phone call to your mother,
or tell the government what you just said on Facebook.

The semiconductor industry is a $300B industry, dominated by global giants like Intel, Samsung, Texas Instruments and Qualcomm, and it’s an industry where the complexity of its products doubles every two years. It costs billions of dollars
to build the factories where the chips are built and millions of
dollars to make the first one of a new design, all so that the chip in
your phone or your car can be cleverer than a mainframe computer was a
few years ago, but only cost a few cents.

None of this would be possible without the computer scientists and
physicists who work in the industry that makes these complex designs
possible. That industry is Electronic Design Automation — EDA — and it
is celebrating its fiftieth birthday this week at the Computer History Museum in Mountain View, Calif.

When the first integrated circuit was designed at TI by Kilby in 1959
design was done by hand. But once the idea was out, a new industry
emerged creating sophisticated software programs running on computers to
help humans create more and more complex designs.

Today integrated circuits are less than 1 square inch in size but are
three-dimensional and have many, many miles of metal interconnect on
them, where every line of metal carries a signal like a wire in your
house, but is thinner than a fraction of a human hair. They can perform
millions of operations per second and store the Encyclopedia Britannica
in your fingertip. And a human mind could not fathom the complexity of
these chips without software programs to control the design and
simulation of the chip before it’s built.

The EDA industry is the group of companies, and brilliant people, who
make the amazing computer brains in the devices we take for granted
every day possible. They build software to model how to turn analog
signals — like your voice — into digital bits. They build simulators
that use physics and maths to model Maxwell’s equations and predict how
electricity is going to move through different materials, at different
speeds. They simulate the memory cells that store data, they predict how
complex logic will work with the different inputs you give it. The
chips being built now have features so small that you can’t use light to
expose them any more (the process is a lot like the old photographic
process) so they use math to adjust how the light will behave and
compensate. It’s rocket science built into software.

But despite being such a profound building block of our modern
electronics, EDA is a relatively small industry. With revenue of $7B,
the industry is dominated by two California Bay Area companies: Synopsys
and Cadence, who work alongside many small, highly innovative
specialist companies to solve the hard design problems (and yes, the
small companies get bought up by the big companies over time). The
industry is small because the number of companies than can actually
afford to design chips is low even though we all use more electronics
every year. But it’s a healthy industry where the leading companies are
growing and generate strong operating margins and where new startups
emerge every year.

And it employs the brightest engineers. Graduates with EECS degrees
(electrical engineering and computer science) from colleges like
Berkeley and Stanford and MIT walk the halls. The executives are all
engineers too because the pace of change of the chip technology is so
fast you need to be able to talk with your customers about what they
need in the language of technology.

Men or women, they’re mostly a nerdy bunch. But tonight, at a banquet to raise money
for the Computer History Museum, they’ll be dressed up and celebrating
their love of one of the most fascinating technical areas you can choose
to work in. And next time your phone, or your camera, or your TV makes
you gasp in wonder think about the software nerds in California who
design the tools, that design the chips, that make your device magical.


Autonomy, Revenue Recognition and the Duping of HP

HP’s acquisition of Autonomy, and subsequent write down of $8.8B amid allegations of fraud by the Autonomy management team, is going to show up as a business school study one of these days, and so it should.

The practice of revenue recognition in software, and how you can be misled by it, is simply not well understood by enough senior management and board members. And as companies shift their software products from license to subscription revenue it becomes imperative that board members do understand it or they can be easily misled.

In the Autonomy case, it appears they did two things that, while not illegal (I think HP will have a hard time proving fraud) are questionable…unethical…short term thinking… pick your poison.

The first is recognizing revenue up front. I interviewed a VP sales candidate from Autonomy a couple of years ago. He was proud of how they were growing their revenue so fast — and I was horrified. They were signing long deals – 6+ years long – and structuring them so they could take the license revenue up front.

As he proudly described to me, he had recently won a very large contract with a global bank to use Autonomy to analyze internal data following the 2008 recession. Sign a 9 year deal, structure it so you take the license revenue up front and maintenance over time (here’s a primer on revenue recognition if it will help), report the revenue on your call as a great deal and never tell anyone that it’s 9 year’s worth, because you are not required to tell them. Pay the sales team commission, the stock goes up, everyone’s happy. Except the analysts who smell a rat but can’t prove it — they will be cautious on your stock.

Software revenue recognition rules are sufficiently complex now that this is not hard to do, it’s all in how you write the contract terms. And because it’s not illegal the auditors, like Deloitte, will not technically cry foul. As Dennis Nally of PwC told the FT last year:

are professional standards out there [and] an audit is not designed
under those standards to detect fraud,” he says, pointing out that
detecting fraudulent behaviour rests on other indications including a
company’s governance, management tone and control systems.”

I agree, it’s all about management tone.

The practice of overly aggressively recognizing revenue up front is not new. Cadence did it for years to inflate their value, and nearly pulled it off.  Before Cadence crashed in October 2008, they were in negotiations with KKR for KKR to take the company private. My network told me (so it’s hearsay) that the deal broke apart on $1 per share. KKR offered $24, Cadence management and board held out for $25. But it was not long afterwards that the board figured out just how much Cadence had been advancing revenue and fired the entire management team.

It takes character and spine to convert your business from up front license revenue to ratable revenue. If you start the business as a SaaS business (like salesforce.com, or FirstRain) your revenue starts out low, but it grows exponentially and you never have to make the switch. But to switch from license to subscription means at some point you have to slow down your growth rate. Both Oracle and SAP are dealing with this right now, and the Autonomy management team must have decided it was easier (and more personally lucrative) to sell to a mug than deal with it themselves.

The second practice reported by AllThingsD is channel stuffing. Again, not fraud but really short term because it creates a future problem every time.

Channel stuffing is selling product on to distributors before they have found a buyer. So this means you sell to your distributor (who is never going to use your product themselves), they pay you, you take revenue and it sits on their shelf until they can find a buyer. This is unforgivable in software.

This practice developed in hardware years ago because the distributor wanted to have the product ready for delivery so they’d buy it from the supplier, but these days, with modern reporting systems, there is no need because your product can sit on your distributor’s shelf while you still own it, and in software there is no reason at all to do it…. unless you are trying to inflate your revenue in the near term and you are willing to bet your revenue will grow fast enough to cover the stuffing.

In the end this is about business judgement and advice. As Reuters headlined a story this morning: In HP-Autonomy debacle, many advisers but little good advice.

Autonomy had the best advisers in the business. They don’t come any better than Frank Quattrone, George Boutros and the Qatalyst crew. I have worked with them on both sides of deals, on my side selling and against me when I was buying, and they know how to develop the case to extract maximum value for the asset they are selling.

Time will tell now whether this colossal acquisition write down was the result of fraud — and so reputations will be destroyed on both sides — or whether it was an overly aggressive tone at Autonomy that inflated the value. But the resulting destruction of value and reputation is the same in both cases.