Google’s Open Source Hardware
Google has been making significant strides in the open hardware arena, with notable collaborations and announcements. One of the major highlights from the past year includes Google’s partnerships with Skywater, Efabless, and GlobalFoundries, enabling individuals to fabricate their custom silicon designs. GlobalFoundries, a prominent pure-play foundry in the United States, joined forces with Google to open-source the Process Design Kits (PDKs) for their 130-nanometer process node.
A Process Design Kit, or PDK, contains crucial design rules, limitations, and various tools essential for fabless designers to create hardware. The Skywater PDK gained remarkable attention on GitHub, accumulating as many stars in a week as other hardware projects like Yosys and Chisel did throughout their entire existence.
Google is extending its open hardware initiatives by working on open sourcing Skywater’s specialized 90-nanometer Fully Depleted Silicon-on-Insulator process node. This unique process employs an insulator to prevent electrons from undesirably tunneling beneath the gate.
In the development from August 2022, Google has open-sourced the PDK for GlobalFoundries’ 180-nanometer process node. This move allows individuals to apply for the fabrication of their open-source designs. Given GlobalFoundries’ significance in the semiconductor manufacturing industry, this announcement signifies Google’s substantial progress in influencing the broader landscape of semiconductor manufacturing.
Google’s View
Absolutely, Google’s not running a charity for hardware design, there’s a savvy business strategy at play. Currently, there’s a lack of a substantial open-source silicon ecosystem for designing data center and consumer applications. When Google creates hardware, like the well-known Google TPU, a significant portion of the design relies on proprietary intellectual property (IP).
This reliance on proprietary IP has contributed to the skyrocketing costs of hardware design. Google aims to tackle this issue by fostering a community of top-notch open-source hardware design projects. The goal is to bend the cost curve, making hardware design more accessible and cost-effective.
Given Google’s colossal scale and the sheer volume of chips they use, any open-source project that reaches a level of quality to replace a proprietary or internally developed component brings not only financial benefits but also performance improvements. So, while Google stands to gain, the broader tech community also reaps the rewards of a more collaborative and cost-efficient approach to hardware design. Win-win!
Strategy
Alright, what’s Google’s big picture strategy for developing this ecosystem, focusing on user-friendliness and accessibility?
The goal here isn’t to produce the next Apple A-series chip or Nvidia H100 super-GPU, especially considering the limitations of a 130nm process fab. Instead, the focus in these early stages is on democratizing chip design, making it more accessible for ordinary individuals to download and utilize tools for creating and sharing their own chip designs. Crucially, this is done without the need to invest in or navigate expensive NDAs for Electronic Design Automation (EDA) software.
The team’s execution plan involves:
1. Opening Up and Unifying Software and Tool Chains:
- Open-sourcing PDKs from semiconductor foundries like GlobalFoundries and Skywater.
- Supporting tools such as OpenLane, an automated flow that integrates various design tools (Yosys, Magic, OpenROAD) to seamlessly guide from design abstraction (RTL) to a foundry-ready GDSII file.
2. Creating Onramps and Incentives:
- Providing tutorials and easily accessible, clone-able projects to encourage adoption.
- Launching a free shuttle fab program, enabling individuals to turn their designs into real chips.
These steps aim to break down barriers, making chip design more inclusive and approachable for a broader range of enthusiasts and innovators.
Projects
Absolutely! While the 130 nanometer process node may not give you the firepower of today’s desktop-class computers, it’s a powerhouse for a different league of projects. The debut of this process back in 2001–2002 birthed the IBM PowerPC 970 chips, a notable piece of tech boasting 58 million transistors packed into a 118 square millimeter chip. These chips found their home in the Power Mac G5, a Steve Jobs revelation in June 2003 with clock speeds ranging from 1.6 to 2.0 gigahertz.
Sure, it’s not the cutting edge by today’s standards, but that’s not the aim here. The 130 nanometer process node shines in the realm of IoT projects, sensor systems, and microcontroller spaces. If you’re after efficiency and low power, you can craft projects that hold their own against contemporary market offerings. Who needs desktop-class performance when you’re building the future of smart, connected devices?
Future
Currently, many of these hardware projects are still in the realm of experimental or academic concepts. However, isn’t that how several major open-source software projects began as well?
Drawing parallels, Linux and GCC, initially considered “toys,” eventually evolved into powerful tools for real-world business applications. Predicting when similar recognition might occur in the hardware domain is uncertain, echoing the unpredictable trajectories of open-source software.
In a SemiAnalysis post, Apple’s efforts to transition numerous non-customer-facing cores from an ARM-based design to RISC-V, adding an intriguing dimension to the hardware landscape.
Moreover, the uptake of open-source hardware design in academia has been impressive. Professors and their teams are embracing the idea of sharing code for global replication, a feat hindered by NDAs associated with closed-source EDA software and PDKs. An example from Brazil showcases a team presenting a hardware accelerator design for the open and royalty-free AV1 video coding format, relying entirely on open-source tools like OpenLane and the Skywater 130nm PDK.
How to get Started?
If diving into this realm piques your interest, Proppy suggests starting with the Notebooks. The team maintains iPython or Jupyter Notebooks — compact, self-contained files allowing the execution of arbitrary Python code. Google’s Colaboratory service enables running these notebooks directly in your browser, eliminating the need for installations.
Begin with the introductory project in the browser, experimenting by tweaking elements to observe their impact on the outcome. This hands-on approach, akin to tinkering with libraries and languages, helps grasp the terminology and lays the groundwork for more advanced projects on the “Build Custom Silicon with Google” website. So, roll up your sleeves and try things out!
References
“Build Open Silicon with Google,” Google Open Source Blog, Jun. 01, 2022. https://opensource.googleblog.com/2022/05/Build%20Open%20Silicon%20with%20Google.html
“Open-source hardware: a growing movement to democratize IC design,” Electrical and Computer Engineering. https://ece.engin.umich.edu/stories/open-source-hardware-a-growing-movement-to-democratize-ic-design/
