Abstract
SADD (Semi-Automatic Digital Designer) is an early knowledge-based system designed to bridge the gap between high-level functional concepts and concrete digital circuit schematics. By employing a "sketchpad" philosophy and a rule-based implementation strategy, SADD translates English-language specifications into chip-level designs, automating the heuristic decision-making process typically performed by human engineers.
Introduction
In the 1970s and 80s, digital design was a labor-intensive process of manual translation. An engineer would start with a functional goal—such as "build a digital clock"—and gradually refine it into a schematic. This process required deep "world knowledge" of available integrated circuits (ICs) and their interconnections. Milton R. Grinberg's SADD system addresses this by acting as an interactive AI assistant that understands both functional intent and technical hardware constraints.
The SADD Methodology
SADD organizes the design process into three distinct phases:
- Specification Acquisition: The user describes the circuit's behavior in English. A phrase-keyword parser interprets these sentences to construct a semantic net of "frames." Each frame represents a functional component and its relationships to others.
- Circuit Design: This phase maps functional frames to real-world chips. SADD evaluates potential ICs using a library of implementation strategies based on three criteria: Prerequisites (mandatory conditions), Eliminators (disqualifying characteristics), and Traits (ranking features).
- Circuit Simulation: Once a schematic is generated, it is tested on a conceptual simulator to ensure it fulfills the original design goals.
Knowledge Representation
A core innovation of SADD is the distinction between "sketchpad" primitives (Functions and Connections) and "implementation" primitives (Chips, Wires, and Signals). This abstraction allows designers to focus on logic before addressing pinouts.
The system stores "prototypical knowledge" using frames, which define the characteristics (ASPECTs) and ports (CONPTs) of generic components. For example, SADD "knows" the general properties of a counter before a specific chip, like the 74161, is selected.
Case Study: The Screensplitter
The system was evaluated using the "Screensplitter," a sophisticated TV video display circuit. The input consisted of a 41-sentence verbal description of the device's functional requirements. SADD successfully processed this input, generating a database of approximately 600 entries and designing sub-components like the Display Character Counter (DCC) that were functionally equivalent to manual designs.
Results and Limitations
SADD demonstrated that it could automatically select correct ICs based on functional requirements, such as choosing the 74161 over the 7493 for specific counting tasks. However, the system faced limitations: the symbolic simulator was not fully integrated for complex debugging, the parser required a specific phrase-keyword structure rather than free-form natural language, and the knowledge base was limited by the number of pre-programmed implementation strategies.
Conclusion
SADD represents a significant early step in automated digital design. By formalizing the heuristic knowledge used by engineers, it proved that high-level functional descriptions could be systematically translated into physical hardware implementations.
Summary updated