Interview with Jim Anderson, CEO of Lattice Semiconductor

People warned me that Xilinx were legally very aggressive and that I could lose everything I owned. Nobody wanted to get involved then. At one point an academic paper about alternative programming techniques on Xilinx devices withdrew some of their work, out of legal concerns. So I kept out of it too.

In fact, I quit working with FPGAs for a long time (it had been my main job, developing and using hardware compilers at a higher level than VHDL/Verilog), so that my previous use of Xilinx and Altera software might become a distant memory and I wouldn't feel that the licensing restrictions accompanying their old software would constrain reverse engineering on a new generation.

I was intending to eventually reverse engineer some of the devices from first principles without using their software. We know that can be done to some extent just by programming bitstreams. Timing and power models are more difficult to reverse engineer, but not impossible and I had some measurement ideas, and I had in mind to additionally etch and photograph the chips as well, or perhaps use focused X-ray tomography. But that never happened; life happened instead.

Times have changed. The FPGA patents from that era have expired. You can build a decent device without infringing, in principle. There's a rapidly growing movement in open source hardware on FPGAs and even ASICs. And people have been reversing Xilinx among others, more confidently than the people of 20 years ago were willing to.

For today I have two words: QuickLogic and OpenFPGA. QuickLogic devices may be small compared with the high end, but they are explicitly supporting open source toolchains, better than Lattice's passive tolerance. This is a breath of fresh air, and I hope they are successful. OpenFPGA is actual open source FPGAs.

I feel this is a mentality change, albeit slow, that is starting to happen. We still don't have source code for Intel MKL, or all of OpenVINO and it bit us in the ass several times, so much I'm really trying to find ANY OTHER SOLUTION and pay the perf hit, if I can own and debug it. I'm not even asking for github or even publicly available tarballs, but 'I paid I want the source and I want to be able to look inside to be autonomous'. Make it OSS without community involvement, for all I care, but dammit let your paying customers look and debug. Let us automatize, let us own the tech.

I'm willing to personally donate 10k a year to projects like these to liberate myself from crappy vendor tools that I can't improve.

Open your stupid bitstream formats FFS.

Amen. The size of these closed source toolchains is often enormous. I wonder how many folks are not content with using relatively large Windows point-and-click software to program relatively small form factor hardware.

Open implementations which acquire enough traction become open standards. I'm doing block-level CPU simulation work for my dissertation, and there's pretty much a wall of abstraction where my work has to stop before it explodes with licensing + compatibility landmines. A rock solid tooling ecosystem I could target for FPGAs and then eventual fabrication would be the dream.

You aren't alone. Most people who work with FPGAs hate the toolchains.

The entire FPGA industry and to a larger extent the entire semi industry in plagued by this problem.

This is a deep cultural issue.

Look for example at NVidia and their dumb secrecy policy around drivers (just the first example on top of my head, examples abound).

Secrecy and making your customers pay for tools is dyed in the DNA of these people and it is going to take a very long time for them to understand the economic advantage of openness.

It took MSFT almost 30 years to finally understand the lesson that giving away dev tools for free (e.g. Visual Studio) was a competitive boost.

Why aren’t there any open FPGAs or at least more Open FPGA tools? Why are vendor FPGA tools so huge, terrible and hard to use? Open FPGA tools are better and the way to go! etc.

In one of Dr. Cutress’ questions he states and then asks: > … they're just hard to develop for. You need to know how to use them before you use them , which sounds like the wrong way to learn how to code! What exactly is Lattice doing to kind of ease that transition for people who may understand software, but are kind of new to the hardware?

Why is this? What can Lattice do?

It’s because one does not “code” an FPGA, they design a microchip and that process involves a lot more than writing code. It is pretty hard to do right. FPGA is on the spectrum of VLSI microchip implementation technologies. That’s why they compare FPGA’s to ASICs in the interview.

Don’t believe me? Look at this article: https://www.tutorialspoint.com/vlsi_design/vlsi_design_fpga_...

So in general the progression is FPGA -> GA (Gate Array) -> Standard Cell ASIC -> Full Custom ASIC

FPGA development is microchip development. It requires a great deal of tacit knowledge, experience and a background in electrical engineering and digital design and a number of other skills to do well.

In order to develop open tools for these devices most of the devs in the community would have to have a lot of this knowledge and experience and most if not all of the FPGA vendor’s IP on all of their devices in order to work on an open tool chain. In the end an open tool chain will do little to alleviate all the pains of the chip design process anyway so it really isn’t worth the effort to make open tools. It is way easier to just use the vendor’s tools once you know what you are doing. Most of the tools have gotten much better in the last 20 years.

That is why you will not see open tools for anything but the smallest and simplest devices that can be reverse engineered.

IMO, the best thing lattice can do for someone who doesn’t have any experience with the technology is to point them to the Lattice Partners program to find an expert to help.

I am probably not making any friends here by saying this, but I welcome any counterpoints we might argue together here.

> One of the things with FPGAs - I get a lot of feedback for is that they're just hard to develop for. You need to know how to use them before you use them, which sounds like the wrong way to learn how to code! What exactly is Lattice doing to kind of ease that transition for people who may understand software, but are kind of new to the hardware?