Paolo Nenzi's Friends

Paolo Nenzi's Groups

Discussions about design tools, algorithms and co-design techniques for semiconductor package design.

June 30

At 8:28pm on July 5, 2008, Theodore Omtzigt said…: Paolo:

Any luck on getting a network of packaging folks interested to discuss packaging problems and the solutions? I have been tracking the network over the past six months and the problem is that nobody seems to use the network for any serious discussions. The thought that we can use this network as an international water cooler hasn't been picked up by many. Any ideas how we can create that realization? The fact that you and I are talking about creating a packaging group to me is such a powerful concept. We would never have connected wasn't it for the concept of a social network given our geographical locations. We are both interested in creating solutions for complex problems and that transcents geography but only if we can create/bring together a group of folks that live and breath this stuff AND that are willing to share their ideas and questions with a larger group. If you have any ideas how we can create that larger group, let me know.

Looking forward to your insights,

Theo

At 2:09pm on April 17, 2008, Theodore Omtzigt said…: I invited Al but no response yet.

On MTL and ngspice: I have been thinking about the 'general' problem of taking a piece of C code and bolting a C++ data structure underneath. The 'architecture' of the code should simplify but to get there might be too steep.

Maybe a better approach is to extract the basic architecture of the code and how it accomplishes its tasks and to map that onto an abstract layer of operators. The goal here is not to prepare the actual coding but to identify universal themes that appear to be true for the circuit simulation category.

Do you have such an analysis for ngspice?

At 3:12am on April 3, 2008, Paolo Nenzi said…: Yes, I think Sandia would like to check reliability in radiation environments.

On the field effects: spice circuit simulators deal with lumped elements and thus developers/researchers tried to put as much as they can into this model. In deep submicron technology one can improve the device models using non-quasi-static approximations, for an example see the BSIM4 device model (berkeley), so yes, some modfication is necessary.

I do not think we will move to 3D simulation on full-chips because the complexity of the problem. Spice is already slow!

Full 3d electromagnetic simulation can be useful to characterize very small and simple parts of the circuit or some integrated sensor.

I think that 3d substrate simulation can be used to build a thermal map of the chip and a "noise map" based on switching activity of transistors.

I see a tremendous importance of physical co-simulation in mems systems, where you have sensors (EM, chemical, etc, mechanical), condition circuitry (RF, low noise tolerant, analog electronics), control circuitry (low power, high density, noisy digital electronics) and actuators (mechanical, etc.).

At 12:14pm on March 23, 2008, Theodore Omtzigt said…: Another question: circuit simulation will become more important for the deep submicron chip technologies. Sandia may be interested in hard radiation problems, but RF and low power electronics make field effects important. This creates a cross over point between regular Kirchoff laws and field effects. I am trying to figure out if this is a multi-physics problem where the field effects change the device models used in the Kirchoff model, or whether this moves circuit simulation into a new realm of looking at the circuit in its 3D physical substrate form.

At 8:57pm on March 22, 2008, Theodore Omtzigt said…: Paolo: the Xyce research is experimenting with parallelization of circuit sims through iterative simulation. My impression from Robert was that it is working very well. I don't understand the architecture quite yet but it is something that I want to get to the bottom of. I have uploaded a couple of presentations from NACDM-2004. One of them compares Kundert's sparse with Tim Davis's KLU: it shows dramatic improvements of KLU over sparse 1.3.

We are working on a generic programming library for matrix math that will try to encapsulate a framework for delivering the main algorithms. Given the fact that it is a generic library it can wrap around existing data structures but it would be better to move the matrix and vector data structures to the native MTL versions since it would allow us to reuse any improvements we make. Connect with Peter Gottschling who is also on this network to get a copy of the MTL to look at. It would be great if we can work together to get MTL-4 into ngspice and make it work well.

The papers I uploaded have a very nice historical description of the issues in circuit sims. If nothing else, replacing the sparse direct solver with KLU will dramatically improve simulation performance.

Theo

At 3:36pm on March 21, 2008, Theodore Omtzigt said…: Paolo:

We have been on a quest to benchmark different solvers. After the discussion with Robert Hoekstra from Sandia, we would like to include an EDA like application like Spice. So we could use your help in this regard:

1- can you help us identify the solver used in NGSpice?
2- can you help us generate matrices that are used with in NGSpice?

We are working on a separate solver abstraction and it would be good if we could feed it with the same matrices that NGSpice is creating.

Is this something you can support?

At 2:06pm on March 9, 2008, Theodore Omtzigt said…: Paolo:

I would love to learn more about ngspice, the roadmap, the things you want to improve, and particularly, the performance of the solver. Stillwater is working on iterative and direct solvers. It is my understanding that SPICE uses a direct solver. When researching circuit simulation solvers I see activity at Sandia but I can't tell what specifically they are doing and the technology they are using. Nor can I figure out if that technology is akin or different from the SPICE world. Do you have any insight in this?

At 12:42am on March 8, 2008, Paolo Nenzi said…: Thanks Theodore for your invitation. Hope I can give contributions to spice discussion. Are you planning to work on circuit simulation ?

At 12:12pm on March 7, 2008, Theodore Omtzigt said…: Paolo:

Welcome! I am glad you joined. I am looking forward to good SPICE discussions. As a matter of fact I am going there right now!

Theo