Podcast: Play in new window | Download
Subscribe: Apple Podcasts | RSS
Welcome back, Andreas Olofsson of ZeroASIC!
- Andreas was on the show back in 2015 (ep 254) talking about the Parallela, a crowdfunded parallel calcuation board by his then-company Adapteva
- What is enabling more open source to happen?
- Unit economics really impact silicon designs
- Open source effects have been having a positive effect on the industry. Andreas maintains a meta repo of 400 tools.
- Fewer fabs than 2008, mask sets still expensive
- Semiconductor singularity
- Andreas is deep into the world of “chiplets”
- This is the basis of his new company ZeroASIC
- Before he started that he was a program manager at a little outfit called DARPA
- Andreas focused on lowering costs, with the idea that 3 people should be able to design a chip
- He worked under Bill Chappell, the director of the Microsystems Technology Office (MTO)
- OpenROAD
- POSH
- CHIPS
- TinyTapeout
- Andrew Kang UCSD
- Chiplets
- We had Ming Zhang on to tlk about ZGlue, but that was a slightly different architecture
- What is a chiplet?
- AXI on chip
- SERDES
- Types of interconnect
- Organic
- Types of output
- SIP, Chip, SOM
- ZeroASIC is Andreas’ latest company
- They started by releasing a Silicon compiler project
- New thing is take system customers and build them an ASIC
- Optimizing speed and cost
- Mostly targeting aerospace and defense
- Try it out yourself on the ZeroASIC emulation page
- Their main processor is a Quadcore RISC V
- There are no off the shelf chiplets
- eFabric Active Interposer
- Defining a standard
- arm made a standard called amba
- AIB from Intel was opensourced
- Getting external contributors (hardware vs software)
- LatchUp – Fossi
- Personal passion drives people to contribute
- You’re really buying a datasheet from a big company
- Loading the design to AWS
- Check out the ZeroASIC openings
- Read about how ZeroASIC is democratizing chip making
The higher architectural levels of today’s proprietary and open source platforms suffer from antiquated technology at their foundations. Your priorities are different from mine. You have been producing different variations on the same old themes for less money while I want something newer, better, more advanced and standalone. I put efficiency and capability over speed. While we have to use our older and more primitive platforms (I am talking about Windows, Mac, Linux Android etc machines) to design and build that better platform, we have to realize that such things as GPUs, Parallela and the like are mere appendages off the current platforms and are not platforms in themselves. The new platform must run standalone.
The collapse of profitable opportunities for “new standalone platforms” in software caused by open source is going to repeat in the hardware space if you are not careful. Just think about it. For open source you now have a choice between Linux, Linux, Linux and of course, Linux. It’s old technology that depends on mountains of library software to glue it all together and much of it is interpretive.
No (useful) open-source platform is fully open source. CUDA/nVidia is not a “platform” because it is a mere appendage off the Microsoft/Apple/Linux monolithic platforms. And, yes, Linux cannot be the embodiment of that new advancement because it is built on antiquated technology. Both Arduino and Raspberry Pi are based on proprietary chips. The existing platforms are needed to get to the next platform but they will never be the next new platform. It will have to be built from the ground up. You should look at the tradeoffs between proprietary and open source and maybe chose a hybrid approach.
Pros:
1. Open to more good ideas into the designs.
2. Get more and cheaper labor because of willingness to work for recognition rather than money.
3. Potential for funding on philosophical grounds.
Cons: 1. Lack of profit motivation. (My labor is worth something and I need to pay my bills)
2. Most brilliant private contributors don’t want to give away their best ideas.
3. Vulnerable to more bad ideas in the designs. Too many cooks spoil the broth.
4. The masses are trained in antiquated technologies and status quo acts as quick sand.
5. Better technologies often fail because of lack of support and understanding by the ignorant masses.
6. Professional lives are wasted by having to learn how to piece their own antiquated open source parts together.
Common downsides of both proprietary and open source:
1. Bullies often get their way while geniuses are stomped into the mud.
2. Propaganda often very effective in misleading information.
3. Competition from many other alternatives. Look at how many Linux OS’s there are.
Common upsides of both proprietary and open source:
1. Can cut through red tape with small teams.
2. Superior technologies can speak for themselves.
3. Propaganda often very effective in misleading information
4. Platform can serve as an excellent learning tool. Can use system to educate and train new developers.
5. The world benefits
I have spent much of my life planning for and designing that new and advanced standalone platform since the mid 1980’s. My key tools for creating this new platform include the C programming language, Verilog, programmable logic devices, various CAD software packages and the in-house proprietary software I have written. I use MS Windows as my primary cross development platform. I could use your admirable IC experience to make the processor technology into ASIC realities.
Out of character Chris kept interrupting as if he had a list of questions to get through and felt the need to turn questions into an incoherent monologue at times. I would have liked to hear what the guest with years of experiences had to say . Preferably “ what have you been up to since we last spoke , what have you learnt , tell us about your new company , and where do you see the industry going – in that order , over to you “