You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
A couple of former Cadence Design employees with experience architecting the Spectre APS and Spectre RF circuit simulators have created a new company called Cigma and a new parallel SPICE circuit simulator named VSIM1. This first product VSIM1 is focused on netlists with up to 10,000 transistors and simulates results up to 15X faster than other parallel circuit simulators by running on a single 16 to 32 core workstation.
A future product dubbed VSIM2 claims to work on netlists with up to 1 billion transistors by running on a distributed cluster of shared-memory computers with up to 512 cores total.
Accuracy claims include a 130 dB dynamic range, and Verilog-A transistor models can be compiled to run faster than hand-coded C models.
If this company wants a quick exit strategy then Synopsys would be the first choice since Synopsys has acquired the most SPICE circuit simulators and could use something newer to replace HSIM and XA, while Mentor Graphics would be a likely second suitor because they don't have a parallel SPICE tool yet. Cadence likely wouldn't be considered as an exit since the founders just left that company.
Jim was the founder and CEO of SDA Systems, which became Cadence Design Systems. He was founder and GM of the Cadence Custom Division where Spectre was developed. Jim led analog and mixed-signal R&D at Motorola and National Semiconductor for 20 years, and has 19 patents in chip design. He is also a recipient of the Kaufman Lifetime Achievement Award, is an IEEE fellow, and has an MSEE from University of California, Berkeley.
Yes, Cigma has an impressive pedigree, however the real big step to any EDA startup is finding your first customers and then showing the world what your tool is really capable of. Right now, Cigma appears to be at the pre-customer stage. We've seen other EDA startups over-state their potential and fall quite short of published expectations, so time will tell if they really have an advantage over the very crowded market of established players.
If this company wants a quick exit strategy then Synopsys would be the first choice since Synopsys has acquired the most SPICE circuit simulators and could use something newer to replace HSIM and XA, while Mentor Graphics would be a likely second suitor because they don't have a parallel SPICE tool yet. Cadence likely wouldn't be considered as an exit since the founders just left that company.
Synopsys already has more (analog) simulators than it knows what to do with - somebody told me 9 - and HSIM/XA is a fast Spice (not a high-accuracy Spice) which accelerates pretty linearly when parallel processed. The threat for Synopsys is that this takes away HSPICE/FineSim licenses. If it supported Verilog-AMS it could definitely take on CustomSim/VCS-AMS, since Synopsys only have the one offering there and it's not fully compliant with the standard.
The main problem with this space is that analog designers hate changing tools so your market penetration is going to be really slow if that's all you can do. Also, this is the one area where there is an open source tool that's probably as good (Xyce - 27x on 256 cores).
Seems unlikely Mentor would buy another "analog fast Spice" after BDA. However, Cadence did buy Antrim and just shelved it because it was an order of magnitude faster than Spectre.
simguru,
We'll just have to wait and see for the first public comments on direct benchmarks of VSIM1 versus anything else on what the simulation speed benefits are.
The real challenge for a new SPICE tool is getting foundry support. Without support from the foundries they are dead in the water. So that is what I would look for, in addition to customer support, is an endorsement from TSMC. Remember, right before BDA was acquired by Mentor they were a TSMC Partner of the year.
The main problem with this space is that analog designers hate changing tools so your market penetration is going to be really slow if that's all you can do. Also, this is the one area where there is an open source tool that's probably as good (Xyce - 27x on 256 cores).
And with free and/or cheap licenses, you can always parallelize with using Amazon cloud machines or some other cloud. Simulation shouldn't be on fixed assets any longer... net throughput may well be highest with a slow free simulator and lots of cheap Amazon clones.
Amazon do have some folks looking after that - I went to a half-day event a couple of years ago. The nice thing for small companies is that the Cloud is elastic and you only pay for what you need when you need it - e.g. simulating extracted net-lists before tape-out.
Another thing you can do in the cloud is very large simulations -
