The engineering discipline of specifying, architecting, implementing, and verifying physical computing and electronic systems — from printed circuit boards and mechatronic assemblies to FPGAs and full-custom silicon — spanning requirements capture, architectural trade-off between performance, power, area, and cost, register-transfer-level description in hardware description languages, synthesis and physical implementation, and exhaustive pre-fabrication verification, since unlike software a shipped hardware error cannot be patched and a mask respin costs months and millions.
Semantic Classification
Content
Definition
Hardware design is the discipline that turns a specification into working physical electronics. Its scope runs from board-level design — schematic capture, component selection, PCB layout, signal and power integrity — through programmable logic on an FPGA, to application-specific integrated circuits and full System-on-Chip designs integrating processor cores, memory controllers, and accelerators on one die. Across all scales the governing trade-off is the same quartet: performance, power, area (or board space and bill-of-materials cost), and schedule.
Digital design practice centres on the register-transfer level: behaviour is expressed in a Hardware Description Language such as VHDL, Verilog, or SystemVerilog, then synthesised by electronic design automation tools into gate netlists, placed and routed, and checked against timing, power, and manufacturing rules. Because a fabricated error cannot be patched — a mask respin at an advanced node costs millions of pounds and a quarter of schedule — verification consumes the majority of engineering effort. Constrained-random simulation under UVM, emulation, and increasingly Formal Verification (model checking and equivalence checking, mainstream since the Pentium FDIV recall made the cost of escapes vivid) together aim to prove the design correct before tape-out.
Hardware design is also a standards-governed activity: interface and memory specifications from bodies such as JEDEC (DDR, LPDDR), PCI-SIG, and MIPI define the contracts a design must honour to interoperate. And it is the physical substrate of every embodied system in this graph — a robot is co-designed hardware and software, its sensing, actuation, compute, and power architecture set by hardware-design decisions long before any control code runs, which is why robotics curricula list hardware design among their core requirements.
Current Landscape
-
EDA consolidation and AI: Synopsys, Cadence, and Siemens EDA dominate tooling; reinforcement-learning-assisted floorplanning and placement now ship in production flows. Google’s macro-placement method — introduced as a 2020 preprint, published in Nature in 2021 and named AlphaChip in September 2024 alongside a Nature addendum and a released pre-trained checkpoint — has been used across TPU v5e, v5p and Trillium and Axion CPU blocks, though its reproducibility remains publicly contested (no independent replication on open benchmarks as of 2026).
-
Open-source hardware: RISC-V has made processor architecture a commons; open toolchains (Yosys, OpenROAD) and open PDKs (SkyWater 130 nm) allow fabricable open-source silicon, with chiplet standards (UCIe) opening multi-die integration.
-
Domain-specific silicon: the end of Dennard scaling pushed differentiation into accelerators — AI training and inference chips, video codecs, network processors — making hardware design a competitive weapon for cloud and automotive firms that once bought commodity parts.
-
Verification frontier: formal methods expand from block-level proofs towards security properties (information-flow, side-channel freedom), while shift-left co-design ties hardware models to software bring-up ever earlier in the schedule.
Sources:
-
https://deepmind.google/blog/how-alphachip-transformed-computer-chip-design/
-
https://github.com/google-research/circuit_training/blob/main/README.md