Podcast: Play in new window | Download
Subscribe: Apple Podcasts | RSS
Welcome Brian Faith, CEO of QuickLogic!
- Past guest Tim Ansell introduced us to Brian, from their work together on the open toolchain.
- They met at a tradeshow and Brian declined the first time, only to be convinced later.
- QuickLogic IP licensing
- Brian attended OR Conf in Bordeaux, where they were watching talks and excited by future growth of users of the open toolchain.
- Resisted for a year
- Brian started at QuickLogic during the “schematic era” (when FPGAs were designed using graphical schematic of logic blocks)
- Previously their toolchain
- Worked with early versions of Synplicity, but later switched to using Mentor Graphics Precision
- There was no bundled simulator
- Proprietary Place and Route (P&R)
- The new QuickLogic approach is Symbiflow
- It’s also about the software engineer
- A community member ported NuttX to the platform
- What did QuickLogic give up, in order to use the Symbiflow toolchain? They had to publish the spec of the bitstream
- What could you do with the spec of the bitstream? Why is it secret? Apparently, due to history and a generally closed off ecosystem in FPGAs.
- QuickLogic is targetign selling to software engineers, not just FPGA engineers. This has become much easier with python targeting FPGAs (LiteX, Migen)
- Software users will help enable more “mass customization”
- Making software designs into silicon
- Open Hardware Group
- RISC V
- Global Foundries at Munich
- The Artic Pro 2 will be built on the Global Foundries 22FDX, which is their 22 nm process
- Hardware/Software partitioning
- They’re building a test chip
- QuickFeather
- SensiML is the web-based machine learning toolset. The team came from the Intel Curie group.
- SensiML was bought by QuickLogic at the beginning of 2019, but they still offer services for chips outside the QuickLogic portfolio as well.
- Chris doesn’t think a threshold detect algorithm would be up to the task in many cases.
- QuickLogic and SensiML are sponsoring a Hackster contest targeted at projects that will help prevent climate change.
- You send in your sensor data, SensiML gives back models you include as a “black box” algorithm
- The web interfice allows you to dial in performance algorithms. You can also update the data/model later if you want to tweak based on new data or different parameters.
- There is an example data set on github using a PM2.5 sensor
- QuickLogic Open Reconfigurable Computing (QORC)
- Size of the model depends on perfomance dialed in on the website
- The models are set to run on on Cortex-M4, specifically the EOS S3
- TensorFlow lite for microcontrollers
- APIs for convolution
- eFPGA = embedded FPGA
- In the case of the EOS S3, it’s roughly equivalent to 1000-2000 LUTS
- USB in the FPGA without a dedicated (hard) USB core can do USB 1.1 full speed data speeds.
- Videos and instructions
- Building a proof of concept
- Community edition of SensiML gives you enough access for entering the contest, trying out models at home (non-commercial).
- If you are developing a commercial product, SensiML has commercial subscription prices (Chris thinks they’re reasonable, relative to hiring an FPGA/DSP engineer)
- Removing the gyro using SensiML
- Wrist worn wearables for applications like remote control
- Industrial applications
- Consumer is still a focus
- Gerry Roston talking about data and monitoring for large scale auto manufacturing facilities
- They are targeting many of their classic customers in the Automotive / MIL / Aero industry, as well as new ones. They are avoiding the server / datacenter industry.
- QuickLogic licenses things like their IP blocks, memory blocks, math blocks for people to design into future silicon.
- If you licence IP from QuickLogic (fabric), you will also be able to use Symbiflow for your silicon product.
- Interested in learning more and giving it a try? Check out the Hackster contest
- EOS S3 page
- Great video tutorials
ben says
Unclear about the “e”FPGA… what is the difference between the EOS S3’s “MCU + eFPGA” and a traditional FPGA with hard-IP ARM (ex. Microsemi SmartFusion2, Xilinx Zynq, etc)?
For example, I assume the EOS S3 is NOT instant-on.. but rather SRAM based and must be configured at power-on?
The Quicklogic website doesn’t simply show what the difference between the various Quicklogic antifuse FPGAs… Eclipse,EclipseII,EclipsePlus,QuickRAM,PolarPro,PolarProII, pASIC3..
Finally, is there a list of soft-IPs that are available either for free, or for purchase (ex. 1553)?