Self-healing concepts involving fine-grained redundancy for electronic systems

Abstract

The start of the digital revolution came through the metal-oxide-semiconductor field-effect transistor (MOSFET) in 1959 followed by massive integration onto a silicon die by means of constant down scaling of individual components. Digital systems for certain applications require fault-tolerance against faults caused by temporary or permanent influence. The most widely used technique is triple module redundancy (TMR) in conjunction with a majority voter, which is regarded as a passive fault mitigation strategy. Design by functional resilience has been applied to circuit structures for increased fault-tolerance and towards self-diagnostic triggered self-healing. The focus of this thesis is therefore to develop new design strategies for fault detection and mitigation within transistor, gate and cell design levels.
The research described in this thesis makes three contributions. The first contribution is based on adding fine-grained transistor level redundancy to logic gates in order to accomplish stuck-at fault-tolerance. The objective is to realise maximum fault-masking for a logic gate with minimal added redundant transistors. In the case of non-maskable stuck-at faults, the gate structure generates an intrinsic indication signal that is suitable for autonomous self-healing functions. As a result, logic circuitry utilising this design is now able to differentiate between gate faults and faults occurring in inter-gate connections. This distinction between fault-types can then be used for triggering selective self-healing responses.
The second contribution is a logic matrix element which applies the three core redundancy concepts of spatial- temporal- and data-redundancy. This logic structure is composed of quad-modular redundant structures and is capable of selective fault-masking and localisation depending of fault-type at the cell level, which is referred to as a spatiotemporal quadded logic cell (QLC) structure. This QLC structure has the capability of cellular self-healing. Through the combination of fault-tolerant and masking logic features the QLC is designed with a fault-behaviour that is equal to existing quadded logic designs using only 33.3% of the equivalent transistor resources. The inherent self-diagnosing feature of QLC is capable of identifying individual faulty cells and can trigger self-healing features.
The final contribution is focused on the conversion of finite state machines (FSM) into memory to achieve better state transition timing, minimal memory utilisation and fault protection compared to common FSM designs. A novel implementation based on content-addressable type memory (CAM) is used to achieve this. The FSM is further enhanced by creating the design out of logic gates of the first contribution by achieving stuck-at fault resilience. Applying cross-data parity checking, the FSM becomes equipped with single bit fault detection and correction.