Heterogenous Integration in High Volume: Now and Looking Forward
Bibliographic record
Abstract
AMD is an industry leader in the use of advanced packaging to enhance its products. Over the past decade, this has included volume production of 2.5D technologies including silicon interposer, organic RDL interposer (FO-RDL), and embedded fan-out bridge (EFB); hybrid bond (SOIC) 3D logic-on-SRAM stacking, and cost-effective chiplet re-use with multi-chip modules. The recently released AMD Instinct MI300 AI chip sets a new standard for what can be achieved from heterogenous integration with 3.5D packaging 3D hybrid bond logic die stacks are connected together and to eight total eight- or twelve-high HBM DRAM stacks using a 2.5D silicon interposer, resulting in a module with massive compute capacity and the near memory capacity and bandwidth needed to support it.This presentation will begin with a brief review of the history of AMDs advanced packaging engagements and the technical incentives for investing in specific technologies. It will then transition into a deeper look at the AI-specific trends that resulted in 3.5D packaging technology. These include compute performance doubling faster than every two years, memory bandwidth doubling every two years, networking bandwidth doubling at only a slightly slower pace, and total power consumption continuing to rise. How heterogenous integration benefits each of these areas will be discussed, along with some of the challenges it introduces. The presentation will conclude with a look at how these trends scale into the future and where AMD is investing its efforts in next-generation technologies. Continuous improvement in 2.5D and 3D capabilities through pitch scaling, increased stacking height, and incorporation of additional features such as high density capacitors remains critical to meet next-generation targets. The evolution of silicon process will also factor into the way next-generation modules are designed, as AI workload power densities strain against the limits of what even advanced liquid cooled thermal solutions can support. The extremely large size of AI modules, their power consumption and growing off-package IO bandwidth requirements necessitate innovations in the rest of the packaging ecosystem as well, as SMT warpage, power delivery, and high-speed IO all have fundamental issues meeting current scaling trends. These challenges ensure there are many important problems for the advanced packaging community to solve in the decade to come.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.003 | 0.003 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.006 | 0.013 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.002 | 0.003 |
| Insufficient payload (model declined to judge) | 0.025 | 0.005 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".