1. Introduction
The designer's CAD system has a perfect digital representation of the board but in transferring the product data to the manufacturer's CAM system much is "lost in translation". Not all useful information is transferred. Much of what is transferred is contained in free format text f iles or drawings, which must be retyped or redrawn manually in the CAM system. This results in wasted labor, longer cycle times and worse, production errors.
The state of the data transfer CAD to CAM is dismal, unworthy of a great technological industry. What is worse is that it barely improving; the quality of data is scarcely better than it was in the 1990's. This article analyzes why progress is so slow and suggests a way forward.
2. Computational geometry is difficult
Experience shows that new geometric applications are initially plagued by tricky bugs. This is not because the programmers implementing them are particularly incompetent or sloppy. It is because this type of programming is very diff icult.
This diff iculty is recognized in the literature. "The Algorithm Design Manual" [1] states: Implementing basic geometric primitives is a task fraught with peril... There are two different issues at work here: geometric degeneracy and numerical instability... Geometric applications can be made robust by writing special code to handle each of the special cases... Expect to expend a lot of effort if you are determined to do it right... The diff iculties associated with producing robust geometric software are still under attack by researchers. Or from Computational Geometry in C : There is no easy solution to the fundamental problems faced here, an instance of what has become known as robust computation. There are several coping strategies...
Each new image format will initially be plagued with bugs. The PCB industry knows this very well. A bug in the CAD output or the CAM input of an image f ile will often not be noticeable in CAM and result in scrap. This explains why the industry is reluctant to take on new image formats. This reluctance is well illustrated by the following example. Some datasets contain the data both in Gerber and in the newer ODB++ format. These datasets often contain a readme. txt file stating: BARE BOARDS MUST BE FABRICATED WITH GERBER, DRILL AND IPC-356 NETLIST PROVIDED. BOARDS ARE NOT TO BE FABRICATED FROM ODB++ FILE.
This does not indicate that there is anything intrinsically wrong with the ODB++ format. On the contrary, the ODB++ is included because it may contain useful information. It does indicate, however, a concern about the reliability of the newer ODB++ software.
3. RS-274X as a 2D image format
The RS-274X format is simple, compact and unequivocal. It is easy to interpret. It describes an image with very high precision. It is complete: one single f ile describes each production layer. It is portable and easy to debug as it uses printable 7-bit ASCII characters. The mix of D codes, G codes and parameters has developed historically and may not be very elegant, but it has all the necessary features: areas, definable apertures, pos/neg exposure etc.
A well-constructed RS-274X f ile precisely defines the PCB image data and the functions of the different image elements. The image and drill layers of the most complex design are being transferred flawlessly in RS-274X. RS-274X is an excellent 2D image format for PCB applications. Most RS-274X software is very robust and very reliable. This is the very rational basis for the popularity of the format.
