The New Golden Age of Computer Architecture: Trends, Challenges, and Opportunities

The article begins with a recap of basic computer and server concepts, referencing earlier posts on system architecture and server hardware.

It then summarizes a 2019 Communications of ACM paper by John Hennessy and David Patterson, highlighting the emergence of Domain‑Specific Architectures (DSA) as the dominant trend for the next decade, the importance of open ISAs (e.g., RISC‑V) as a "Linux for processors," and the possibility of agile hardware development using ECAD tools.

Key historical laws—Moore's Law, Dennard scaling, and Amdahl's Law—are revisited, explaining how the slowdown of Dennard scaling and the end of Moore's Law limit traditional performance gains and motivate new architectural approaches.

The article examines the transition from CISC to RISC, the rise of VLIW/EPIC, and the eventual dominance of RISC in the post‑PC era, noting that x86 market share is declining while RISC‑based designs are proliferating.

Current challenges are discussed, including the energy inefficiency of speculative execution, the impact of security vulnerabilities such as Meltdown and Spectre, and the constraints imposed by thermal design power (TDP) leading to the "dark silicon" problem.

Future opportunities are explored: improving high‑level language performance, leveraging DSA for specialized workloads (GPU, TPU, FPGA), adopting open ISAs like RISC‑V, and using domain‑specific languages to achieve better parallelism, memory efficiency, and precision scaling.

The article also covers agile hardware development workflows—simulation, FPGA prototyping, ECAD layout, and tape‑in—showing how design cycles can be reduced from months to weeks.

In conclusion, the authors argue that while traditional scaling laws have ended, the combination of software innovations, open architectures, and specialized accelerators will usher in a new golden age for computer architecture, with RISC‑V likely becoming the dominant ISA.