Lambda scaling comes from the old MOSIS SCMOS technologies. Since it is fundamentally built into magic, I keep the lambda concept; however, it is not particularly relevant to anything. I set lambda to 0.01um for the convenience of the units. The minimum manufacturing grid of the 130nm process is 0.005um. The internal units are set to the minimum manufacturing grid, which is why there are 2 internal units per lambda. I find it generally unhelpful to work on the (default) lambda grid, which is why I usually just run the command "snap internal" to work directly on the minimum manufacturing grid.