As engineers, all of us probably had to take at least one economics course in college. It was usually one of those curriculum “requirements” that we agonized through, amused only by the discovery that Laplace transforms were useful for something besides jumping between frequency and time domains.
(If you’re one of our few non-engineers, don’t let that last line scare you away. We’re going to be talking about economics here – supply and demand and competition and giving stuff away for free and the system collapsing in on itself and leaving us in poverty after agonizing years of … Oh, sorry. Got carried away there.)
In our typical economics class, apart from doing Laplace transforms, we had some scenario with two guys running two factories – probably both making widgets. The first guy, we’ll call him “Bob,” can make a widget for $10 and sell it for $20. The second guy, whom we’ll call “Sanjay,” amazingly makes widgets for the same amount and sells them for the same amount. In order to gain advantage, and depending on whether widgets exhibit price elasticity and a bunch of other economics class blah blah that the professor might have mentioned (while we slipped out the back of class to see how far our FPGA-based prime number generator lab had gotten in the last hour, and then sneaked back in just in time to get our homework assignment), Bob and Sanjay have several options to try to make more money. They can raise or lower the price they charge for widgets, they can try to reduce the amount it costs them to make widgets, or they can try to improve their widgets. (We recommend that they add widget Wi-Fi, or at least upgrade them to USB 2.0, but it all depends on what widget technology was like when you were in school. You may have been upgrading your widgets from 45RPM SP to 33RPM LP…)
Bob and Sanjay’s economy is manufacturing-based. Widgets are physical objects that must be individually manufactured, packaged, sold, and shipped, and there is a non-trivial cost involved in producing each one. Life is comparatively simple in manufacturing-based economies. Sure, we have all the variables like cost of goods and shipping and packaging and… some other stuff that may have come up on a day of class we missed, but all those variables can be handled by a couple of Laplace transforms.
More often today, however, our economy is becoming based on a different type of commodity – intellectual property. Luckily, this is one we engineers are all familiar with, since the things we have to sell to the world are our ideas. It’s probably obvious that an idea is not a straightforward thing to sell. You can’t package it up in a cardboard box, put a barcode on the bottom, and have people slide their debit cards at the checkout stand. Normally, to sell ideas, we have to contrive some way to get value. A lot of us sell our ideas as services. Our employer pays us for our time, and we reward them with our ideas. They, in turn, usually re-sell those ideas by reproducing them in some manufactured embodiment – like an MP3 player, which they then sell in a manufacturing economy, where the rules are back to well-behaved again. Get those Laplace transforms tuned up!
Unfortunately, some ideas don’t get bundled with tangible physical objects. Software, for example, is a big problem. Since software can be duplicated and distributed essentially for free, the normal rules of a competition-based economy start to break down. (For those of you still following along with the math subtitles, you plug in that big zero as a denominator somewhere, and the equations just all go to heck.) Since every competitor can keep dropping their price per copy down to essentially zero, and since the barrier to entry for a competitive product is also near zero, everything – including the funds available to develop and improve your product – also starts to head rapidly toward the big dollar-sign-naught.
Outside of economics class, this effect is a real-world problem for companies that develop and market software. These companies have struggled with a plethora of different business models, trying to find a way to preserve the value in their intellectual property. In EDA, for example, most products used to be sold on a perpetual-license basis – often supplemented by support services. Without going into too much detail (and to save you from having to skip several paragraphs to check on your FPGA-based prime number generator again), the perpetual license model is fraught with peril. Software gets pirated, customers don’t upgrade to the latest versions, your support staff gets spread out supporting old software and helping work around bugs that have been fixed for years, your revenue is unpredictable, your competitors have an in every time you release something new… OK, maybe that’s too much detail already.
The best solution (so far) to this conundrum is to sell software as a service (SaaS). Instead of buying a copy of a specific version of your software to use forever, your customers buy the rights to use the services of your software for a specified time. SaaS has a number of benefits for both customer and supplier. It makes the software companies’ IP delivery value-scalable. It makes revenue more predictable, and it eases support. From the customer side, ownership is much easier, and the supplier is constantly incentivized to improve and update the product in order to preserve their revenue stream.
SaaS is a proven, successful model. It provides a sustainable, defensible way to build a business extracting value from software IP. As we just noted, however, SaaS was really motivated by the fact that software broke the traditional manufacturing economy model with its near-zero manufacturing and distribution cost. What does that have to do with FPGAs?
Let’s now consider the concept of Moore’s Law. As we all know full well, Moore’s Law has had us delivering exponentially more electronic hardware capability per unit cost for upwards of four decades now. If we re-arrange the cost/value equations we can see that the exponential decrease in cost, over time, will make the cost of electronic hardware approach zero. As cost approaches zero, our hardware economy starts to look a lot like a software economy. We start to bump into the choppy seas where we have to consider a new business model to extract value from our IP. If we follow in the footsteps of software, one viable option is likely to be hardware as a service (HaaS). HaaS is not a new idea, and cloud computing servers, set-top boxes, and mobile phones are already essentially shifted to an HaaS model.
As in software, the key enabler in selling as a service is programmability. By making hardware programmable, FPGAs enable generic, inexpensive hardware platforms to be distributed, and complex services to be delivered, controlled, and upgraded in the field. FPGA-based platforms can be extremely generic, and the specific services — comprising hardware and software — can be installed and managed after the fact.
As engineers, why should we care about this? Because we need to think differently when we design an FPGA into our system. The FPGA is not just another component adding cost to the BOM, sucking power from the supply, and doing it’s pre-conceived job. The FPGA in our system beckons us to think differently about the product we’re developing – to morph that product towards a service – a long-term commitment to our customer to evolve and improve – a longer life in the field for hardware and a shorter life for problems and limitations. If we use the FPGA as a lynchpin from which we evolve our product into a customer experience or a service, we are preparing ourselves to be more competitive in the harsh reality of the new hardware economy.