CMOS

SKY’s the Limit with the SKY130 Open-Source PDK

By Rocket55 |

The open-source ecosystem sparks innovation, lowers barriers to chip design and paves a new path for global workforce development

by SkyWater CTO Steve Kosier

Three years ago, SkyWater, Efabless and Google teamed up to release the industry’s first open-source foundry Process Design Kit, or PDK, known as SKY130 — based on SkyWater’s volume 130 nm CMOS technology. The partnership gave designers worldwide free access to chip design technology to create new, manufacturable designs. While the focus of this effort was largely on revolutionizing technology realization for all — some exciting social developments also emerged.

Interested in an update, we sat down with SkyWater’s Chief Technology Officer Steve Kosier to talk about the SKY130 PDK — curious to learn how it’s helping the innovation ecosystem globally while introducing chip design skills to an organically developing talent pipeline — with free educational tools for anyone interested in developing a skillset as a chip designer.

The huge cost of chip design stifles innovation and cripples education

Semiconductor design costs grow exponentially with each new node. To use commercial tools and design a production-ready 130 nm chip, you can expect to spend about $5,000,000. Migrate to 65 nm, and you’ll need $20,000,000. If we move to the leading-edge technology, 5 nm, the bill can be as much as half a billion dollars. For 2 nm is even more expensive at $725,000,000. Obviously, such figures put chip design out of reach for an individual user.

There are other barriers to chip design too. You need Electronic Design Automation, or EDA tools, non-disclosure agreements, intellectual property agreements, access to fabrication and testing. For many aspiring chip designers, organizations, and creators of new products — these hurdles can be insurmountable.

In educational institutions, access to chip design tools that target manufacturable process flows can be limited and unaffordable — leaving students with theoretical knowledge but lacking practical experience. Furthermore, the inability to share information between institutions, reference prior designs by other researchers, or afford wafer fabrication makes educating and attracting students to semiconductor careers difficult.

What if the cost was closer to $0?

Enter the value proposition for open-source PDKs. Open source tools are free to all, which eliminates the cost to get started — unleashing creativity and making chip design available to more people. The Google-sponsored Open Multi Project Wafer, or Open MPW, Shuttle Program attracted designers from academia, commercial organizations and beyond, enabling market-relevant designs to be fabricated by people who traditionally would not have had the opportunity to design a chip.

In the first MPW run in 2020 there were 42 submissions and about 60% of the designs were submitted by non-chip design experts — demonstrating a significant untapped interest generated by putting design capability in the hands of non-traditional users. In subsequent MPW runs, the submission rate grew to 162 in 2021 and 418 in 2022.

How did it all come together? SkyWater licensed the technology to Google who then released the PDK and funded eight MPW shuttles. This was in collaboration with our partner, Efabless, who delivered the engineering platform, design aggregation, and developed key intellectual property.

No-cost tools and low-cost fabrication is democratizing high-end silicon design

Our partner in this effort, Efabless, provides ChipIgite which consists of Caravel, a pre-designed carrier-chip along with an automated open-source design flow called OpenLane, making creating a chip easy and affordable. This rapid path to prototyping and low-volume production is ideal for educational institutions, startups, and individuals who want to design market-relevant products.

Co-founder and CTO of Efabless Mohamed Kassem explains, “The number of institutions of learning on SKY130 is rapidly increasing. More than 50 universities are using the SKY130/eFabless platform for classwork, capstone projects and research projects — but we don’t know the total number as it’s free and open for all to use — which has been the main lubricant to gearing up this success so quickly.”

In some cases, students who take such classes use slots on a ChipIgnite, which will be fabricated by SkyWater. ChipIgnite is scheduled in April, June, September and November so that students get silicon back the following session to test and demonstrate projects with working implementations.

All users appreciate:

  • No legal agreements — All students, researchers, and professors access foundry technology, chip reference designs, EDA tools and design flows that are completely open-source. There are no NDAs or other legal requirements. All content can be downloaded from git repos. URL https://skywater-pdk.readthedocs.io/en/main/
  • No export control restrictions — This makes it easy to provide access for all people.
  • Freedom to collaborate — Share design files across teams and organizations, including GDS final layout files.
  • Rapid design — Complete a design in a full chip within a single course or cycle of learning by leveraging reference designs and automated flows. By using the Caravel SoC on SKY130, it’s possible to complete a focused design goal quickly.
  • Tiny Tapeout – This program allows a designer to go from idea to chip design in minutes. A student can now do a design for $50. An eight-year-old created a chip using this innovation.
  • New research-focused tools — A new NIST and Google Nano Accelerator and Open Cryo PDK built on the SKY130 platform delivers the scale of open IP and multi project wafers to the research community.
  • MPW shuttle program – The Google-funded MPW shuttles helped get this program started. While those initial eight shuttle runs have concluded, programs like Tiny Tapeout and ChipIgnite are available for designers to design and manufacture affordably.

 

130 nm CMOS is a proven workhorse for mixed-signal applications and ideal learning node

Opening development on this widely used and commercially relevant node is ideal because it has a large pool of existing IP given its maturity — something that’s valuable for seasoned and novice chip designers alike. The more advanced the node, the more design rules — and the longer it takes to validate and manufacture a chip.

One university professor who started teaching with SKY130 moved to a 16 nm node platform for a semester. She has returned to SKY130 saying that it was too hard for first-time chip designers to complete a chip using the 16 nm alternative. SKY130’s simplicity and compatibility with early education is essential to giving students a practical experience where they can learn and successfully create a chip.

A talented community is building the future in the “Innovation Tail”

The greatest chip production volumes serve leading edge applications such as smartphones, PCs, memory, etc. While they have huge volumes — they also have very few designs. On the other end, there is a very long tail where many millions of products exist. This is where future innovation is happening. It makes sense to make technology accessible to all so that we can improve what’s there and grow the knowledge base of the world.

 

Since SkyWater released its open source SKY130 PDK there are now more than 6,400 members in the Slack interactive discussion workspace and membership continues to grow! There are also 100+ channels including #analog-design, #courses and #riscv. The best part is that anyone can join here.

On Google-sponsored shuttle MPW-8, the fastest ever design creation rate emerged. A remarkable 147 unique market-ready manufacturable designs were submitted in a lottery for free fabrication. Although only 40 designs could be chosen for fabrication, the work and designs are still available for all to reference, use and improve.

There were many interesting designs submitted including sleep apnea devices, a flu detector, smart garden and image detection at the edge. These innovative designs speak to the whole purpose of the ecosystem: people can come up with their designs and share them with the world.

We need more people interested in semiconductor careers

One of the challenges facing the semiconductor industry is a shortage of talent. As technology continues to advance, the demand for skilled professionals in semiconductor design and fabrication is growing. To address this issue, we need to capture the attention of younger students —those entering high school — to increase awareness earlier. In addition, we need to emphasize the value of semiconductor careers and make education and training more accessible.

Open-source initiatives can play a pivotal role in workforce development by providing learning resources and hands-on experience opportunities for aspiring semiconductor professionals. We couldn’t do this before because of the large barriers to entry. Now with open-source, there are no questions. We can create textbooks, share resources, and create programs of learning that benefit everyone.

Start designing today at Efabless.com or Tiny Tapeout

The open-source revolution in semiconductor design is transforming an industry once shrouded in exclusivity into one that welcomes innovation from all corners of the globe. By lowering the cost of design tools, enabling affordable fabrication, and fostering a global community of chip designers, the open-source ecosystem is democratizing high-end silicon design.

In fact, SkyWater recently announced the availability of a new Cadence open-source PDK and reference design which will be available in the Cadence® Very Large-Scale Integration Fundamentals Education Kit. The kit teaches students how theories and concepts can be applied in the design of simple logic circuits and in the physical implementation of a simplified microprocessor. This is exciting news, and we anticipate more announcements like this soon.

It is inspiring seeing everyone using and making improvements to the SKY130 PDK. It takes everyone working together to improve the community and advance the state of open-source tools. If you have ever wanted to design a chip, now you can. It’s never been easier to get started.

I would encourage you to investigate the SKY130 PDK. Who knows — your design could solve a key problem, move rapidly to volume production, and improve the world!

ReRAM NVM – Coming Soon for your Embedded Design

By Rocket55 |

SkyWater and Weebit Nano have been working together for several months following the announcement of our agreement through which SkyWater will take Weebit’s ReRAM Non-Volatile Memory, or NVM technology, to volume production.

We sat down with Ross Miller, VP strategic marketing & business unit at SkyWater, and Eran Briman, Weebit Nano’s VP of marketing and business development, to understand more about the partnership and hear about what’s next.

Can you give us a quick overview of Weebit Nano and what the company provides?

Eran: Weebit Nano is a developer of Resistive RAM, or ReRAM — a new type of NVM. Up until now, the standard for this type of memory has been flash technology, but with the emergence of new applications, skyrocketing data movement and storage needs, and tightening cost and power constraints, the industry is looking for a new type of NVM.

Weebit was founded to provide NVM that is higher performance and lower power than flash and has a long list of other advantages. A key consideration from the outset of the company was ensuring the technology would be commercially accessible, so our ReRAM is based on the most common materials and equipment used in fabs today. This makes it low cost and easy to integrate into existing flows and processes.

What are Weebit and SkyWater doing together?

Ross: SkyWater helps our customers bring new ideas to market with differentiated technologies, materials and processes. To this end, we partner with companies like Weebit that have bold ideas, solid technology, and the ability to have a big impact on our customers’ designs. Weebit ReRAM is an innovative emerging technology that is at the forefront of NVM. Now it is reaching maturity and approaching volume production.

SkyWater will initially offer Weebit ReRAM to customers as embedded NVM IP on our 130nm CMOS process. The technology can also be scaled in the future to a number of other platforms such as our 90nm and carbon nanotube CMOS technologies. The Weebit team has a great deal of commercialization experience, making the process quite straightforward as we move toward production.

Can you tell me a little bit about the applications where you see Weebit’s ReRAM being a good fit initially?

Eran: Since nearly every electronic product requires NVM, and ReRAM enables semiconductor memory elements to be much faster, lower cost, more reliable and more energy efficient than those using flash, target applications are extremely broad and diverse. Some initial applications where we see ReRAM being adopted are industrial, automotive and aerospace & defense.

Ross: SkyWater’s 130nm process, where we’re first rolling out the Weebit technology, is a sweet spot for a broad range of mixed-signal designs like analog, power management, sensors as well as rad-hard designs. Because ReRAM has such low power consumption and integration flexibility, we see a lot of interest in the near-term in IoT and general mixed-signal / ASIC designs.

Can you explain why ReRAM is a good fit for some of the vertical market segments? Let’s start with aerospace and defense.

Ross: Aerospace and defense applications are among the most demanding, requiring extremely high levels of precision and accuracy. ICs for aerospace and defense also have unique requirements for robustness and reliability in harsh environmental conditions. Since these products must often last for years – mostly without maintenance – longevity is another key trait.

Eran: Weebit ReRAM provides excellent endurance and retention even at high temperatures, retaining data for up to 20 years at 175 degrees Celsius. It can also withstand 350x more ionizing radiation than flash. If you’re designing a product today with flash, it’s going to require extra design for redundancy or shielding, and this isn’t needed with ReRAM. Our technology also has advantages over other emerging NVMs like MRAM such as cost, access time and immunity to electromagnetic fields.

What about industrial and automotive applications?

Ross: Both industrial and automotive ICs must be designed for safety and longevity, as well as robustness against vibration, humidity, extreme temperatures and other harsh environmental conditions. In rechargeable-battery applications like electric vehicles, there is also a need to withstand electromagnetic interference. The key with safety-critical industrial and automotive technologies is using proven, solid technologies. SkyWater’s proven analog/mixed-signal process technologies are a great fit for both industrial and automotive ICs, and we’re able to provide a high degree of customization needed to enable highly differentiated products.

Eran: The number of chips across the car and industrial environments continues to increase. Most of these chips require NVM, whether it be for trimming, code storage, or data logging. Key considerations are fast boot and code execution, handling frequent updates, and responding instantly even during power loss. For the growing number of small, low-power sensors that are often in remote and inaccessible locations, embedded NVM must also have ultra-low power consumption, and the performance to enable the shortest possible active power usage. Along with the excellent retention and endurance that Ross mentioned, this is what ReRAM offers.

Where do you see ReRAM going in the future?

Eran: As discussed, there is a broad and growing opportunity for ReRAM to replace flash across a broad range of applications. A bit longer term, one of the exciting opportunities is in neuromorphic computing where ReRAM isn’t acting as a traditional NVM, but also as a computing element. Today’s AI uses artificial neural networks to simulate brain function, but with neuromorphic computing, we’re talking about true brain inspired AI systems. Since a ReRAM cell physically resembles a biological synapse, it has functional similarities as well. This means that emulating neural networks with ReRAM consumes orders of magnitude less power than today’s neural network simulations.

Weebit has been working for several years with a number of global academic institutions toward ReRAM-based neuromorphic computing. A couple of years ago, together with our development partner CEA-Leti, we were the first to demonstrate ReRAM based spiking neural networks, and now we’re working on the next generation of this. Industry-wide, we’re starting to see neuromorphic developments moving from academia into commercial companies – both established and startups. With this move, ReRAM-based neuromorphic computing might be closer than people think.

Ross: ReRAM fits with SkyWater’s vision for the future as we look toward technologies that will enable differentiation and advanced performance the market demands. Consumers have a growing expectation of more capabilities and functionality at the same cost, so on-chip integration continues to increase. The natural evolution of this is toward hybrid monolithic integration, combining for example a MEMS structure or photonics device on top of a CMOS wafer. Or layers of carbon nanotube-based CMOS with layers of NVM.

Because ReRAM is a back-end-of-line, or BEOL technology, it can be integrated at various points in the fabrication process. And because it can be used as a computing element as well as an NVM, there is the possibility to increase density and bring more device-level capability to designs. Weebit ReRAM is a rich building block for hybrid architectures that can be flexibly integrated in new and interesting ways, making it a really exciting option for innovators as they look to bring new ideas from concept to reality.

SkyWater has a lot of experience in this area with the work we’ve done with MIT on Darpa’s 3DSoC architecture. That work showed the viability of this type of paradigm-shifting solution. Now with Weebit ReRAM we can make it a commercial reality.

Where do you stand in the technology transfer process, and when can customers get started?

Eran: Since announcing our partnership we’ve been working together with Weebit’s R&D partner CEA-Leti on transferring the Weebit ReRAM technology to SkyWater’s production fab. This process is moving forward smoothly as planned and once it is completed, qualification can begin..

Ross: As soon as the initial qualification is complete, SkyWater customers can have the confidence to tape-out their designs with ReRAM. Customers should reach out to us now to learn more about how ReRAM can help differentiate their solution for their specific application.