Overview of ASIC Chip Types, Characteristics, and Applications

ASIC (Application Specific Integrated Circuit) chips are custom-designed integrated circuits tailored to specific electronic system requirements, offering optimized computational efficiency for fixed algorithms and finding applications in AI devices, cryptocurrency mining, printing, and defense equipment.

Hardware-wise, ASICs are built from silicon, gallium phosphide, gallium arsenide, or gallium nitride, and structurally consist of external storage, power management, audio/video processors, and network IP cores, with modules capable of hosting multiple ASICs to meet diverse needs.

Classification by customization degree: full-custom ASICs (highest performance, longer design time, up to 9 weeks per unit, up to 8× the compute of semi-custom designs), semi-custom ASICs (standard cell libraries with some custom logic, including gate-array and standard-cell variants), and programmable ASICs (FPGA and PLD families).

Gate-array sub‑types: channel, channel‑less, and structured arrays, each differing in transistor placement flexibility and routing methods.

Functional classifications: TPU (Tensor Processing Unit for machine learning), BPU (Brain Processing Unit by Horizon Robotics), and NPU (Neural Processing Unit for deep‑learning workloads).

Key advantages: reduced area due to elimination of redundant logic, lower power consumption per compute (≈0.2 W per unit vs. 0.4 W for GPUs), high integration, and lower cost (average ≈ $3 per chip, with potential further reductions at volume).

Key disadvantages: long design cycles, high dependency on specific algorithms, and risk of market obsolescence due to rapid AI algorithm evolution.

Product examples: Google’s 2016 TPU, IBM’s 28 nm TrueNorth brain‑inspired chip, Intel’s Xeon ASIC line, Stanford’s neuromorphic ASIC with 9,000× speedup, and various emerging startups applying ASICs to security, autonomous driving, smart appliances, and medical devices.