The LTSpice IV Simulator: Manual, Methods and

It is an honor to write a preface for Gilles Brocard. I appreciate his work writing this book and hope you benefit from
his labors.
LTspice has been fun to write. It let me implement a number of numerical methods that make LTspice better than
traditional
SPICE programs: a new numerical integration method, node reduction, a native circuit element that behaves
like a power MOSFET, and new time step size control to name a few.
The biggest recent advance in LTspice was when it went multi-threaded in 2008. We found it easy to distribute the
computations over multiple cores but challenging to make the simulation actually run faster. The problem was that
the LTspice object code had been so optimized (much had already been implemented in optimized assembly language)
that it didn't take very many microseconds per timestep and that was a short time compared to how well one
can synchronize multiple threads. That's when we developed a means to dynamically adjust each threads' cache size
to stochastic cool the threads to keep the work load spread evenly. Another important technique introduced at that
time was code generation that generates an assembly listing optimized for your circuit. Then that code is assembled
and linked by LTspice for execution. This self-authoring code is generated typically every few seconds during the simulation
to help your circuit execute close to the theoretical flop limit of a modern CPU. That's why LTspice IV is fast.
But all this is for a purpose. I believe SPICE has impacted mankind more than any other simulator. Writing a better
SPICE is important. LTspice offers you the ability to rapidly prototype your designs so that you understand them better
and even develop intuition.