 So, yeah, thank you for having me and I'm happy to be here to give a short intro about lack of open-source silicon ecosystem. And so my name is Joanne of Rosine, online I go by Prope. I joined Google like in 2011, so a little bit more, 12 years ago. I've been doing a developer relationship there. So I'm a developer relationship engineer, meaning that I'm focusing on improving the developer experience for a variety of Google developer products. I started working on some of the cloud developer products of Google first, then I worked a little bit on Android 2, and more recently I've worked on IoT and hardware things. This year, or last year, I joined a new team at Google called the hardware toolchain team and my presentation is about like what this team is doing and how we are trying to build an open-source silicon ecosystem. So my core team mission is that it's to make custom silicon easier to build for everyone at scale, just like software. If you can think of something like, I know many of you are a developer and you are familiar with some optimization file that you can pass to your compiler. And sometimes you go, you max like those optimization possibilities, and there is nothing that you can do in software to optimize further. So our vision is that we'd like you to help you to go to the next step, to help you to optimize your stuff in hardware. And usually the gap between jumping from a software product to a hardware product is pretty big. And we'd like to make it that as easy as just changing your compiler flag. So we're not there yet, but that's like the vision. So imagine that you have an optimization flag and you can like say that you want to optimize stuff in silicon, and then give you the hardware design to optimize like that translation unit. And in order to get there, we found out that there are like missing pieces. So in ICCAD in 2020, like the team founder actually like released a paper there called The Missing Pieces of Open Design Enablements. Enable, he identified like four things that are missing for creating like an open ecosystem in hardware that could strive as much as a software one. So first like open source PDK. So in software world, we take it for granted that we have like some open source SDK and library that we can build or work on. And often like work is consistent remixing like a few libraries that we want to produce something original. Like it doesn't really work like this in hardware, like the very lower level of the stack, which is a PDK, you can think of it as an SDK for hardware. It's a thing that define like the specification of the foundry, of the process of the foundry, the thing that will allow you to manufacture something with that foundry. It's an interface of the foundry. And those things currently require, most of them require an NDA. And so like you can't even start developing your hardware project until you sign a contract with the foundry. The other thing is like, and so we realize that if we want to allow people to do custom hardware, we have to have open source PDK. The other thing is a C contusion. So we take it for granted in software that most of the compiler or debugger or optimization tool are open source. And you can run it on any computer and you can run it in the cloud. Like it doesn't really work like this. It wasn't really working like this in hardware like a few years ago. Like most of the tools were proprietary. You had like proactively licensed cost to get started. And even more important, you couldn't run like this tool really freely on the cloud or on weird architecture. The other thing is that like because there is this bottom of the stack with the tooling and the PDK that were mostly proprietary. It meant that basically all the building blocks that people were building on top of those two were also tied to proprietary solution. So there was like very little reuse of like other people and sharing between people design. And the last thing is that it's, it costs a lot to manufacture silicone. It can cost like tens of thousands and hundreds of thousands of dollars for just to like a simple batch on a very, very old technology. It could cost like as millions of dollars to do something on a more modern process. Meaning that the cost of error is huge. Like if you make a design mistake, like it's gonna cost you like thousands or millions of dollars. And so it's really a priority for people to get into that field because like in order to learn something you have to make errors. And if they say, or could you cost you a huge amount of money, you're not really incentivized to learn this. And so we identify that if we basically provide open source PDK and open source silicone to Chen, if we provide an easy way for people to generate blocks that they can share and reuse. And if we provide them with cheaper and faster manufacturing option, we can create like bootstrap an open source ecosystem that could be comparable with software. And so that's what we try to do with that team. So we partner with a few foundry across like the world. So the first one is Skywater. And we manage like to convince them to open source a PDK for one of their older 130 nanometer technology. So there is like two variants of the PDK. So there is a Sky 1 and more in 30A. That's like a regular CMOS process and there is a Sky 1.30B. Which is as the same process with a rerun cell. So resistive RAM capability added. And that was like two years ago. Last year we managed to convince another fundry, global fundry, to open source their PDK for their 180 nanometer technology. So it was very important for the ecosystem because like we went from one to two. So the gap between zero to one was big. But and a lot of the tooling actually started like supporting like the Skywater PDK. And the gap between one to two was also very big because like all the tooling had to kind of remove the thing that were specific to Skywater to start like supporting another PDK. And like this year there was also something that was very good for the ecosystem that happened. So we had this fundry in Austria called IHP that released their open source PDK, their PDK, an open source PDK for their 180 nanometer technology without talking to our team at all. So before like Google kind of went to this fundry, tried to lobby them or convince them to open source PDK. And we at HP we had like the first example of an organic release of a fundry that didn't consult at all with us. And that started releasing like their technology. And later this year we hope that we'll be able to release like a 90 nanometer PDK for the Sky90 PDK process. And so once you have the PDK, you can run like a torsion on top of it. And so we support like many of those open source tools from that ecosystem. There is tools that we develop ourselves. So like for example, XLS is a 90 level synthesis tool. It's that Google developed in us. And that we make sure that these tools support like the various PDK that got open sourced. We rely for synthesis on the tool called YOSIS. That's very well known also inside the FPGA ecosystem. For place and route, we rely on open road and with a flow called open lane. And for the last step, which is producing a file that you send to the fundry for manufacturing, we rely on magic and Kaliot. And so all these kind of tools can give you an idea on how you can go to a high level description that's really close to source code that describes the functionality that you want to accelerate. So in our case like it's a X file, it's a syntax that's close to Rust that allow you to describe like a algorithm that you want to create hardware for. And then you go to Verilog, which is like something that gets really tied to, it's basically, it can get synthesized into an at least that much like the process that you want to use. And then like this at least get plastered with like the sale on natural silicon dye. And that's like what the left and the death are about. And then you stream that into a GDS format, which is describing the actual polygon that are being sent to the fundry for fabrication. And like all these tools can allow you to like produce this file like using only an open source tool. And once you have the PDK and you have the tool, you still need a way to manufacture those chips. And like I mentioned before, like it could cost like tens of thousands or maybe like hundreds of thousands of dollars to manufacture chips on those old, even on those older technology. So we started for the two years, we've been running like an open MPW Shuttle program that basically come from no cost to the community. Like the community only need to produce like an open source design that's reproducible with the open source tool to unroll into that program. And we've been selecting like 40 projects on each Shuttle round to get manufactured. And so like at the end, like you get like a board like this with little custom silicon chip on it. And then you can like start like validating that the silicon result that you have actually matched like the intent of your design. Like every single chip come with a user area that you see here where you can put your own logic and a little like this white chip on the bottom that can help you like troubleshoot your design. And that's like one of the picture of the wafer that got manufactured on one of the first one. So we've seen an increasing engagement from the community on the Shuttle. So we run like nine Shuttle on this program. And like here you can see like we started with only like 40 projects at then 55 and on the last Shuttle we had as most as 150 projects that got submitted. And so it means that we have committed like a pool of more than 700 design that are fully open source and reproducible with the open source tool. So the tool have changed a lot. So it could be challenging to reproduce like one of the earlier one with like the newest version of the tool. But still the source is available. Like anybody could like look at the project and anybody could try to reuse like the bits that are there. And we've seen engagement from all over the world on this program. So that's like kind of a picture of like the the breakdown of project by country for one of the latest run that we run like GF MPW Zero. And so it's not only the United States we've seen like very good engagement from country in APAC like Japan and India. And currently we are busy with the community trying to bring up like the silicone for the second Shuttle, the MPW 2.1. And here you can see the various example of people like playing with the ball like that silicone is actually working and like kind of posting like the result online. I wanted to showcase like three projects that we've seen on the second Shuttle where we've got like good results. So that one is like from an actual professional like a sub designer from Intel that has like 20 experience in hardware design. And so he did like this little RIS5 microcontroller that's compatible with Arduino. And like it has like an SD RAM memory controller on board it also have a quite SPI like interface. And he was able to get like his chip booting and validate that the silicone was working. And something that he said is that he like he has been in this industry for a long time he wouldn't think that it will be possible for him to kind of build like a custom silicone directly from his own laptop and being able to manufacture it for free. So someone else that I thought was like was bringing up was like GetCat with like the project that he did with their university from the L.D. Burke University. So what they did is that they did a custom FPGA. So they implemented like an FPGA fabric using the Skyweather program. So you see in the case of Skyweather process so it's a tiny amount of loot there. It's only 100 lutes. But the achievement is impressive because here they were able to run like a custom FPGA logic using an open source FPGA tool chain targeting a chip that didn't exist before now running on the FPGA, a custom FPGA fabric. So the amount of things that could fail there in that stack is like impressive the tool chain could possibly not work like the design could be flowed like the silicone could also be problematic and like they were able to validate that silicone and validate and here you see like in the middle they have a little FPGA demo that run on their fabric and that show it working. And so yeah, that was like impressive that they managed to get that working on a brand new open source PDK. And the last one is like kind of a dial well into the narrative of democratizing IC fabrication. So it comes from an online course called Zero to Azix where they actually instead of using like the whole chip area for just one design what they did is that they run that course online without really asking us for any permission. And they just entered like a shuttle and not submitted one design but submitted 16 design from the school from this course into one project slot. And like most of the people on that I think like 12 of the people of the 16 were actually first time designer. They never did a hardware design before and they were going through that course curriculum and at the end like we're able to submit their own design. And so it also show like how the community can if you give like the community like constraint like for example, you can fit like inside that amount of silicone area they will be like their own process and their own learning material and they will kind of adapt to this constraint and try to build something else. And I mean, I think it's very impressive what they managed to do here because they managed to get a lot more people on top of this infrastructure that we set up with better learning experience that we could have built. Okay, so the other thing I want to tell you is to like please, if you're interested in hardware like feel free to join this community. We have a Slack channel with more than 4,000 members and we have like this website called developer.google. And I have a talk about that where I will actually go inside running this tool with you on Friday. So if you're interested, feel free to come there. Yeah, sorry, I went over.