Luke Kenneth Casson Leighton

nmigen™ is a tool for creating hardware, whether for ASICs or for FPGAs. it is not itself an actual language (like Verilog or VHDL), and it is not like MyHDL which allows translation of a limited subset of python source code into verilog. Instead, nmigen allows you - in python - to create HDL constructs, and to mix those in with the full power of python OO techniques: objects, classes, even multiple inheritance, which is sorely lacking in the Hardware world.

this talk will go through the background and origins of nmigen, present a short worked-example, present some best-practices for people wishing to transfer over from another HDL, and cover some of the planned advancements.

The Libre-SOC Project aims to develop a family of OpenPOWER processors, ultimately working towards a full hybrid 3D CPU-VPU-GPU. After two years of development funded by NLnet, the current phase is to develop a Gigabit Ethernet Router ASIC. Funded by NGI POINTER, this ASIC is intended to be capable of running OpenWRT with OpenPOWER-compliant RADIX Virtual Memory.

This talk will provide background, a status update, and go through the incremental development and debugging that is leading up to being able to boot a microwatt linux-5.7 kernel with a buildroot initramfs, under verilator.

Coriolis is a suite of software for taking chip designs and turning them into ASICs (aka VLSI design). It is silicon-proven, having been used with Alliance auto-scaleable nsxlib Cell Libraries to produce ("tape out") NDA-free GDS-II files that resulted in successful working 180nm ASICs, and has been used in ASICs up to 800k gates. The input is HDL (verilog, VHDL, and yosys RTLIL) and the output is 100% complete GDS-II, with IO pads, cells, SRAMs all fully "Placed and Routed", ready for ASIC manufacture.

However unlike traditional P&R software (which may or may not have had, at some point in its development, a scripting language added as an afterthought), coriolis2 is written in a hybrid of c++ and python modules. Layout is done not by a file format that must be loaded by a GUI: layout for each ASIC is actually a python program that, through the coriolis2 python modules imported by that program, reads the HDL, reads the Cell Libraries, and, fully under parametric programmatic control, creates the layout.

This talk provides a demo walkthrough of coriolis2 in action, to produce an actual GDS-II layout including the IO Ring and IO Pads.

The LibreSOC hybrid 3D CPU-VPU-GPU is intended to provide a significant reduction in both hardware complexity, software (driver) complexity and systems integration primarily initially for embedded and mobile environments.

Larrabee or more specifically Nyuzi showed that a software-only "Traditional Vector Processor" architecture makes for a fantastic High Performance Compute Engine that, unfortunately, also turns out to have only 25% the performance/watt of current competitive embedded mobile GPUs. Not only that but SIMD, despite being (seductively) easy to implement by hardware engineers, has been shown to have harmful consequences at the software level (setup and loop end cleanup). A recent patch to glibc6 to add POWER9 VSX strncpy was a whopping 250 hand-crafted assembly instructions, where its equivalent using Cray Vector principles is around 14.

All of this was solved decades ago by Cray Vector designs, and then forgotten. Only now is variable-length Vectorisation being rediscovered and deployed in modern architectures: RISC-V RVV, ARM SVE2 and also Simple-V. This talk therefore goes through the background and concepts behind Simple-V. Thanks to a grant from NLnet, SV will be formally documented and proposed formally as an extension to OpenPOWER, for review by the OpenPOWER Foundation.

The LibreSOC Project is a hybrid 3D CPU, GPU and VPU, designed for use in mass-volume products such as smartphones netbooks tablets and Industrial SBC IoT. As such, user trust and reduced product development costs are both equally important. Both these goal are achieved by providing full source right to the bedrock (Hardware HDL, bootloader, drivers, everything) and deploying strict transparent "Libre" development criteria. The project has EUR 350,000 funding from NLnet under their PET Programme and is actively seeking developers.

Analysis of SoCs (system-on-a-chip) used for embedded, mobile and IoT shows a decade-long disturbing persistent trend that is almost 100% without exception across hundreds of integrated processors: not a single recent SoC with modern performance in the past 10 years can be used 100% effectively without at least one closed source driver. The choices for developers and end-users alike is very stark: go without the GPU, or without the VPU, or without some key critical functionality that would penalise performance or business effectiveness entirely, or compromise on integrity and end-user trust, and reduce product reliability by delivering closed source binary-only drivers.

Even the highly-regarded Freescale/NXP iMX.6 which uses Etnaviv for the GPU, and has a 19-year Long-Term Supply committment, is still burdened by closed source 3rd party binary-only drivers for the VPU. Huge multi-man-year efforts consuming precious FOSS engineer resources that could be productively deployed elsewhere are wasted on reverse-engineering these closed SoCs.

The kicker: unlike a Desktop system where swapping out the Graphics Card is a 15 minute job, you cannot replace the VPU or GPU or crypto-block on an integrated System-on-a-Chip, and you certainly cannot disable the boot-level DRM that prevents booting anything other than pre-approved signed versions of u-boot (Otherwise fantastic Windows 10 ARM laptop products using the Tegra processor quickly became landfill due to the BIOS-level DRM locking that chained right through to applications. applications that could only be downloaded once a Microsoft online account had been registered)

This situation is one that is not deliberate: it's a product of how SoCs are put together. It is simply easier for Fabless Semi Companies to license off-the-shelf solutions (GPU hard macro, VPU hard macro) with their associated closed source secretive drivers than it is to tackle what is effectively, to their perspective, a hindrance to sales and profits: issues faced by end-users who know nothing about source code or the close relationship that source code has to "Right to Repair" and security vulnerabilities.

Unbelievably, then, after more than 10 years of waiting for even just one SoC manufacturer to come out with an SoC that has full source code for its GPU, VPU, bootloader and all other integrated HDL, the solution is to design and bring to market an SoC that does exactly that, developing integrated 3D GPU and VPU capability into the same and providing a 100% committment to provide full source, right to the bedrock: HDL, bootloader, drivers, OS: everything.

This huge project requires concerted community effort and coordination with many stakeholders. Significant care is taken to appraise the OpenPOWER Foundation of the ongoing efforts and enhancements being developed on top of OpenPOWER, with a view to proposing the 3D and Video instruction enhancements to the OpenPOWER Foundation for review.

2,500 people kindly backed the EOMA68 Libre Laptop and EOMA68-A20 Computer Card Crowdsupply Campaign last year. This talk will briefly outline the progress and some of the strange-seeming decisions that have had to have been made. It's also worth noting that at the beginning of the year, Intel! Announced! The World's! First! Ever! Modular! Computer Card! - in reality they're actually about 6th down a long list. We're not worried about them copying the concept, and will explain why during the talk (one hint: Intel backdoor spyware co-processor...)

Also the reasoning behind why the project also includes the creation of an entirely new 3D Printer (the Riki200) and also a new 3D printer Controller PCB will be explained (big hint: cost. Incredibly, an entire 3D printer's components can be sourced in Shenzhen for LESS money than a western-designed and manufactured DuetWifi 3D Controller Board!)

Please note that this replaces the talk by Rik Lempens on SiriDB - Time Series Database

In reaching out to the India Shakti RISC-V team the opportunity presented itself to put a proposal to them of a mobile / embedded RISC-V 64-bit SoC that would meet their requirements: low cost, libre, and suitable for four markets: smartphone, tablet, laptop / netbook and embedded industrial purposes. Six years ago the author attempted to create an SoC pinmux: it took over two months. Learning from that experience and instead writing a python program to represent the pinouts, adding new test scenarios instead took about an hour each, including altering the pinouts to best match the new scenario whilst still maintaining access to functions needed for all other scenarios as well. Assuming the market assessments were correct, the design - which only requires a 300-pin BGA package - can be said to have been proven to successfully meet all four target markets. Crucially with this approach, potential customers can be approached with the preliminary design, for their input and feedback before committing huge sums to design and tape-out actual silicon.

The output of the python program is a simple markdown page that can be used, without alteration, as the Reference Documentation should the SoC ever be created. In addition to the pinouts, the source of the design inspiration and guides utilised in the design has also been documented, so as to provide not just Reference Schematics and parts that are easily available in the Shenzhen / China markets, but also a logical justification for the actual choice of interfaces in each of the target usage scenarios. Ultimately, it helps explain why low-cost embedded and mobile-class processors are designed the way that they are.

Also included is links to numerous sources of BSD-licensed and compatible LGPL-licensed hard macros (VLSI / VHDL) such as the ORSOC Graphics Accelerator, so that, if utilised, apart from the DDR3 hard macros (which the Shakti team plan to implement under a libre license) the entire SoC's full hardware-design source code is completely available under Libre Licenses. With absolutely no proprietary firmware required whatsoever it has the potential to be one of the world's first ever mass-volume mobile-class quad-core 28nm 2.5ghz embedded SoCs that will be entirely libre right down to the bedrock.

Bunnie Huang's "Guide to Shenzhen" is absolutely invaluable, but it assumes that you know exactly what you need to get, before you actually get it. If this isn't clear, for an example of exactly what the problem is, try an Internet search for "Laptop trackpad supplier". Did you find one? Did you find a source of trackpads for use in that laptop that you're thinking of designing? Or did your search queries instead get completely and utterly swamped with thousands upon thousands of aliexpress pictures of the EXACT SAME Apple Trackpad for a 13in Macbook Pro?

What about LCDs? Do you even know what type of LCD is commonly available right now? Were you aware that MIPI is Cartelled? Or that the people telling you that eDP is "The Future" are trying to push a much higher-power-consumption design at you and they forgot to invite China-based Manufacturers to the closed-door meeting. How can you actually find which LCDs are commonly manufactured and available in Shenzhen, so that you can begin the design right now before you take that expensive trip abroad.

And what about components? If you've looked on taobao or aliexpress, you'll know that many of them are rip-offs, there's absolutely no datasheets, the vendors make pittance so they're extremely unlikely to answer your questions... so what actual use is this resource, to anyone?

This talk will help answer the paradox of bridging the gap between Shenzhen market prices and the Western and Libre mindsets for hardware design, allowing attendees to take proper advantage of the lower pricing of Shenzhen-sourced parts, and to avoid the nightmare scenario of designing a PCB only to find that its parts went End-of-Life in the Shenzhen markets many years ago, and that the last remaining parts available must be bought from Digikey and IMPORTED (back into China from where they originated!) with a whopping lead time and huge import tariffs of up to 40 percent.

About 5 years ago it hit me that Corporations are never going to take responsibility for Software Libre compliance. Full Software Libre compliance actually affects their bottom line, and it's the responsibility of Directors to pathologically maximise profits. In fact, selling eco-conscious products that can be upgraded year-on-year, throwing away perfectly good components in the process, also adversely affects a Corporation's bottom line.

Competing with existing profit-maximising Corporations on "Software Libre Compliance" is not a strong selling point: nobody who isn't a programmer actually really cares, and they're not knowledgeable enough to upgrade the software anyway - they'd rather just throw the whole thing away and get a new one, in the naive hope that the new OS will somehow be better, cleaner, and free from the malware attacks that slowed down the perfectly good hardware they just discarded into landfill.

People are mainly influenced by money (saving it), cool-factor, convenience and fear. So the trick is to come up with a strategy that plays on all these four things. The EOMA68 Project thus was envisioned as a way to help people save money (because they can upgrade just like popping out a "Memory Card", they can now push a button and pop out the "Computer Card"), as well as help reassure them on personal security. Sending the laptop back for repair, you don't want the people in the shop to copy or rifle through your private data? No problem: pop out the "Computer Card" before sending it off... and incidentally you can pop that Computer Card into a spare base unit and carry on working.

Almost as an incidental side-line, then, the products being developed are Software Libre Compliant right from the ground up (not that the end-users really need to know that, as it's not a major selling-point to them). Some of the products being developed can even be RYF Certified (are FSF-Endorseable) - again, the average end-user doesn't care about this. However, Software Libre developers will know what's under the hood, and are being invited to participate in bringing these products to market. The first main products will be two EOMA68-compliant CPU Cards (one with an FSF-Endorseable Ingenic jz4775, the other with an Allwinner A20 dual-core ARM Cortex A7), a Micro-Desktop "base" and a 15.6in 1366x768 Laptop. The laptop's case is 3D-printable on a standard 200x200mm Mendel90, and uses bamboo laminate for the main panels, to save on plastic.

In short: where standard Hardware Corporations are set on maximising profits year after year, collectively they've left a gaping hole in the market that, with a little creative thinking, can be filled by someone who is willing to take active responsibility for doing things in a conscientious and principled way. The resultant products - which have an expected upgrade lifecycle of at least a decade - just happen to be really cool.

(Please note that this talk replaces "Emulating the Nintendo 3DS" by Tony Wasserka.)

About 5 years ago it hit me that Corporations are never going to take responsibility for Software Libre compliance. Full Software Libre compliance actually affects their bottom line, and it's the responsibility of Directors to pathologically maximise profits. In fact, selling eco-conscious products that can be upgraded year-on-year, throwing away perfectly good components in the process, also adversely affects a Corporation's bottom line.

Competing with existing profit-maximising Corporations on "Software Libre Compliance" is not a strong selling point: nobody who isn't a programmer actually really cares, and they're not knowledgeable enough to upgrade the software anyway - they'd rather just throw the whole thing away and get a new one, in the naive hope that the new OS will somehow be better, cleaner, and free from the malware attacks that slowed down the perfectly good hardware they just discarded into landfill.

People are mainly influenced by money (saving it), cool-factor, convenience and fear. So the trick is to come up with a strategy that plays on all these four things. The EOMA68 Project thus was envisioned as a way to help people save money (because they can upgrade just like popping out a "Memory Card", they can now push a button and pop out the "Computer Card"), as well as help reassure them on personal security. Sending the laptop back for repair, you don't want the people in the shop to copy or rifle through your private data? No problem: pop out the "Computer Card" before sending it off... and incidentally you can pop that Computer Card into a spare base unit and carry on working.

Almost as an incidental side-line, then, the products being developed are Software Libre Compliant right from the ground up (not that the end-users really need to know that, as it's not a major selling-point to them). Some of the products being developed can even be RYF Certified (are FSF-Endorseable) - again, the average end-user doesn't care about this. However, Software Libre developers will know what's under the hood, and are being invited to participate in bringing these products to market. The first main products will be two EOMA68-compliant CPU Cards (one with an FSF-Endorseable Ingenic jz4775, the other with an Allwinner A20 dual-core ARM Cortex A7), a Micro-Desktop "base" and a 15.6in 1366x768 Laptop. The laptop's case is 3D-printable on a standard 200x200mm Mendel90, and uses bamboo laminate for the main panels, to save on plastic.

In short: where standard Hardware Corporations are set on maximising profits year after year, collectively they've left a gaping hole in the market that, with a little creative thinking, can be filled by someone who is willing to take active responsibility for doing things in a conscientious and principled way. The resultant products - which have an expected upgrade lifecycle of at least a decade - just happen to be really cool.