The prospects for analogue neural VLSI.

Area of Science:

• Neuroscience
• Computer Engineering
• Robotics

Background:

• Analogue neural hardware development has transitioned from general-purpose neurocomputers to specialized applications.
• This shift is driven by the unique advantages analogue computation offers in specific domains.
• Understanding the fundamental differences between digital and analogue computation is crucial.

Purpose of the Study:

• To explain the reasons behind the shift towards niche markets in analogue neural hardware design.
• To explore the benefits of pure analogue and pulsed design methodologies.
• To investigate critical challenges in analogue neural machine design.

Main Methods:

• Comparative analysis of digital and analogue computation principles.
• Review of pure analogue and pulsed design techniques.
• Investigation of key analogue design issues: weight storage, on-chip learning, and arithmetic accuracy concerning noise.

Main Results:

• Analogue computation offers distinct advantages for specific applications, particularly in sensor fusion and robotics.
• Weight storage (volatile and non-volatile), on-chip learning, and noise-induced arithmetic inaccuracies are identified as key design challenges.
• Specific areas where analogue techniques are most beneficial are outlined.

Conclusions:

• Analogue neural hardware is best suited for niche markets requiring specialized computational capabilities.
• Addressing design challenges related to memory, learning, and accuracy is vital for future development.
• Analogue techniques hold significant long-term utility in targeted applications within robotics and sensor fusion.