每日轻资讯Daily Notes
从晶圆增长、工艺 AI 到量子材料：今天三个可制造性信号。From wafer growth and process AI to quantum materials: three manufacturability signals.
今天三条信号指向同一个变化：微纳项目的竞争力正在从“能否做出单个样品”，转向“来料是否稳定、过程是否可观察、性能是否能在实验室外可靠复现”。这要求项目团队从第一版工艺包就把材料、数据和验收放在一起。All three signals point to the same shift: micro- and nanofabrication projects are moving from proving a single sample to controlling incoming material, observing the process and reproducing performance outside the lab. Materials, data and acceptance criteria should enter the first process package together.
SEMI 的晶圆统计说明，材料端的增长已经扩展到功率器件和光子等更多方向。SEMI’s wafer data shows materials growth extending into power devices and photonics.
SEMI 报告，2026 年第二季度全球硅晶圆出货量达到 35.73 亿平方英寸，同比增加 7.4%，环比增加 9.1%；增长动力除先进逻辑和存储外，也包括功率器件与光子市场。对微纳项目而言，晶圆供应增加并不等于任意来料都可直接替换，晶向、掺杂、厚度、表面处理和批次一致性仍要写进规格。SEMI reports Q2 2026 worldwide silicon wafer shipments of 3,573 million square inches, up 7.4% year over year and 9.1% quarter over quarter, with demand reaching beyond advanced logic and memory into power devices and photonics. More supply does not make substrates interchangeable: orientation, doping, thickness, surface finish and lot consistency still belong in the specification.
SEMI 的制造 AI 工作坊把焦点放在可追溯、可解释和能跨工序使用的数据闭环。SEMI’s manufacturing-AI workshop focuses on traceable, interpretable, cross-step data loops.
8 月 5–6 日的议程覆盖良率管理、故障检测、Run-to-Run 控制、DOE、设备健康监测与根因分析，并明确指出只服务单个工序的模型难以支撑完整 traveler 决策。对研发样品，最现实的第一步不是追求“大模型”，而是统一样品编号、设备号、配方版本、量测坐标和异常处置记录。The August 5–6 agenda covers yield management, fault detection, run-to-run control, DOE, equipment health and root-cause analysis, while noting that step-specific models do not support full-traveler decisions well. For R&D samples, the practical first move is not a larger model but consistent sample IDs, tool IDs, recipe versions, metrology coordinates and exception records.
NSF 对量子材料的总结再次强调：新奇性能之后，仍要回答规模制造与实验室外可靠性。NSF’s quantum-materials overview returns to scale-up and reliability beyond the lab.
NSF 介绍了量子点、量子阱、量子线、二维材料和拓扑材料等长期研究积累，同时把“如何规模制造”和“如何在实验室外可靠工作”列为仍待解决的问题。对材料器件项目，漂亮的低温或单点结果只是起点，还需要建立厚度、缺陷、界面、环境稳定性与器件性能之间的对应关系。NSF reviews decades of work on quantum dots, wells, wires, 2D materials and topological materials, while identifying scalable manufacturing and reliable performance outside the lab as open challenges. A strong low-temperature or single-point result is only a start; projects need mappings among thickness, defects, interfaces, environmental stability and device performance.
一个工艺观察Process Note
可制造性不是最后一次评审，而是从来料到验收持续更新的一条证据链。Manufacturability is a living evidence chain from incoming material to acceptance.
只保存最终 SEM 图或测试曲线，往往无法解释批次差异。更有效的做法是让每个样品都能回溯到来料批次、关键设备与配方、过程量测、异常处理和最终判据，再用 DOE 或统计方法判断哪些变量真正影响结果。Saving only the final SEM image or test curve rarely explains lot variation. Each sample should trace back to its incoming lot, critical tools and recipes, in-process measurements, exception handling and final criteria, so DOE or statistics can identify the variables that truly matter.
项目准备提醒Project Prep
提交一张“来料—工序—量测—判定”追溯表。Submit an incoming-material–process–metrology–decision traceability table.
至少写清基底规格和批次、样品编号、关键配方版本、量测位置与方法、异常记录、目标值和合格区间。微纳Hub 可据此判断哪些参数必须锁定、哪些量测需要加密，以及下一轮 DOE 应优先改变什么。At minimum, record substrate specification and lot, sample ID, critical recipe versions, measurement location and method, exceptions, target values and acceptance bands. MN Fab Hub can then identify what must be locked, where metrology needs more coverage and which variables deserve the next DOE.