The Changing ASICs Landscape: the Shift Toward Chip Disaggregation

The rapid rise of AI is reshaping the demands placed on silicon. For ASICs, this means workloads are becoming more system-specific while architectures grow increasingly modular and disaggregated. These shifts place greater weight on cross-domain collaboration spanning architecture, packaging, and manufacturing.

For many years, ASICs were designed around relatively stable standards and well-defined functions: video codecs, networking protocols, signal processing pipelines, and so forth. Performance gains largely came from integrating more functionality onto a single monolithic die and relying on process scaling to do the rest.

That model is now under pressure. AI workloads are highly diverse, compute-intensive, and tightly coupled to software behavior and system context. A single “best” architecture no longer exists. Instead, performance, power, and cost are determined by how well silicon is tailored to a specific combination of algorithms, data flows, deployment constraints, and operating environments.

To read the full article, click here