Expertise in Semiconductor Industry │Silicon Photonics│AI Industry│Supply Chain│ASIC│Ecosystem │Food │Travel

Taiwan
這個LV行李箱應該沒辦法托運🤣
2
3
1,684
The ISS orbits 400 km from Earth; the Moon is 384,400 km from Earth and communications take 2,5 seconds; Mars is between 55,7 million to 401,3 million kilometres from Earth and communications between both planets take from 6 to 44 minutes. tspasemiconductor.substack.c…
1
1,159
From 800G and 1.6T to 3.2T: The Real Question Is How Fast Each Lane Can Become The most immediate impact of AI data centers is the relentless increase in Ethernet bandwidth. As the market moves from 400G and 800G into 1.6T, with 3.2T already appearing on the horizon, optical modules cannot continue solving the bandwidth problem simply by adding more fibers and more optical lanes. Per-lane transmission speed therefore has to increase. The transition toward 800G and 1.6T has been built around 100G-class electrical lanes and increasingly 200G-class optical lanes. Looking beyond this generation, the industry’s attention is already turning toward 400 Gb/s per lane. This is why the InP Mach-Zehnder modulator demonstrated by SMART Photonics at ECOC 2026 deserves particular attention. tspasemiconductor.substack.c…
1
3
28
2,384
#Infineon and #SolarEdge expanded their collaboration on solid-state circuit breakers for 800VDC AI data-center distribution. The design combines power semiconductors and electronic protection intended to interrupt DC faults faster than conventional mechanical breakers. tspasemiconductor.substack.c…
1
1
5
1,207
PCB equipment and semiconductor equipment were historically viewed as relatively separate markets, but their technological boundaries are becoming increasingly blurred as high-speed SerDes, CPO, chiplets, and advanced packaging converge. tspasemiconductor.substack.c…
1
4
22
1,804
🌝May the moon stay bright, the wafers stay round—and the yield stay high. tspasemiconductor.substack.c…
Made with AI
1
3
1,227
ABF substrate yield also depends heavily on circuit complexity. The roughly 60% yield mentioned earlier is closer to a relatively simple design. For advanced AI or CPO substrates around 10cm × 10cm, supply-chain feedback suggests yields can fall below 50%. Larger and more complex substrates may be closer to 40%. That changes material consumption dramatically: 60% yield → 1.67× input per good substrate 50% yield → 2.0× 40% yield → 2.5× And much of the scrap has already consumed multiple ABF layers and expensive processing steps. So the real ABF demand equation is increasingly: Area × Layers × Design Complexity ÷ Yield This is why a small reallocation of ABF supply cannot solve the shortage. Poor yield itself is becoming a major source of material consumption.
ABF substrate S/D gap is expected to become progressively worse year over year, reaching a S/D gap of 46% in 2028. Up from just 1% in 2026. Hello ABF substrate department? Yes, please send us more ABF we need it desperately!
3
5
71
6,252
A tightly integrated ecosystem in which wafer manufacturing, advanced packaging, equipment, materials, qualification, and mass production can iterate together over extremely short physical distances — and at extremely high speed. tspasemiconductor.substack.c…
4
1,042
Why Has Mining Indonesia Become One of Southeast Asia’s Largest Mining Exhibitions? Mining Indonesia positions itself as one of the largest international mining equipment exhibitions in Southeast Asia, and there is a structural reason why a show like this can become so large. It sits at the intersection of three major forces. The first is resources. Indonesia is one of the world’s most important suppliers of coal, nickel, and several other mineral resources. That means mining equipment is not being displayed only for branding purposes. There is genuine, large-scale local demand. The second is industrialization. Indonesia is attempting to move from exporting raw materials toward processing more of those materials domestically. Once that happens, mining demand expands into smelters, metal processing, power systems, water treatment, construction, and industrial facilities. The third is infrastructure. Indonesia has a population of more than 280 million and continues to urbanize and industrialize. Mines, roads, ports, industrial parks, power grids, water infrastructure, and data centers all require substantial capital expenditure. tspasemiconductor.substack.c…
1
3
1,076
Laser integration will form another major part of the discussion. We will compare the emerging role of micro-transfer printing with wafer bonding, die bonding, flip-chip integration, and other heterogeneous integration approaches, focusing on the manufacturing implications for future silicon-photonics and Optical I/O platforms. At the same time, multi-core fiber is moving from laboratory demonstrations of transmission capacity toward practical connector engineering. We will therefore examine metasurface-based PIC-to-MCF coupling, polarization and spatial multiplexing, rotational alignment, assembly automation, and why fiber density could become one of the next major physical bottlenecks for CPO. tspasemiconductor.substack.c…
1
15
1,385
Beyond conventional optical interconnects, ECOC 2026 also shows that the role of photonics is expanding rapidly. We will discuss optical computing moving from matrix multiplication toward higher-order tensor operations, fiber evolving from a passive communication medium into a distributed sensing infrastructure, live telecom fibers simultaneously carrying traffic and detecting environmental signals, and PON architectures being explored for distributed edge-AI training. Automotive optical networking, hollow-core fiber, free-space optical backhaul, and emerging quantum-network protocols further demonstrate that optical technologies are beginning to penetrate entirely new system architectures. tspasemiconductor.substack.c…
1
2
9
1,151
晶圓代工龍頭台積電( #TSMC)持續擴大其位於美國亞利桑那州鳳凰城晶圓廠的營運規模,目前正積極面向美國退伍軍人招募高科技人才,以填補該廠區約 2,500 個職缺需求! 隨著新廠逐步推進先進製程與擴建進程,當地半導體技術專業人力的供應成為維持產能擴張的關鍵環節。退伍軍人具備的技術背景、設備維護經驗與高度組織紀律,被視為能有效縮短無塵室與高階製造訓練週期的優質勞動力來源。 此項招募計畫不僅旨在緩解美國本土半導體製造面臨的勞動力短缺問題,亦顯示企業正透過多元管道在地扎根,加速建立當地的半導體供應鏈生態系。 資料來源: 鏈新聞
1
11
2,150
#Renesas 印度封裝廠進入Yield Optimization:印度開始真正累積「Production Learning」 日本Renesas Electronics⁠參與的CG Semi Sanand封裝廠,在今年7月開始生產後,已經進入 yield與throughput optimization階段。G1最大年產能約3億顆;更大的G2正在興建,目標在2027年底至2028年初將整體能力推向約40億顆/
1
6
1,548
AI公司開始直接找Memory原廠:HBM採購權正在從GPU Vendor往Hyperscaler移動 9月23日韓國產業報導指出,隨著HBM與server DRAM供應持續吃緊,AI infrastructure業者正更積極直接與memory manufacturers談長期供應,而不是完全透過GPU/ASIC供應商間接取得memory。
3
28
2,755
True….
BREAKING: New study finds companies using AI mostly hire in senior roles and cut entry-level jobs
3
1
9
2,598
I believe AI can change the world....
卧槽??一个被人类算法学家研究了几十年的问题,Opus 5.5 居然又抠出了一个新解法? @ValsAI 让 10 个 Claude Agent 去解决一个人类研究了几十年的经典问题:能不能把最短路径算法再往前推一步? 这类问题难在哪? 因为这类算法,已经被几代计算机科学家研究得很透了。 重要的是这不是什么优化一下代码就完事儿的事情,而要从数学上证明即使数据规模越大,你新算法确实比以前更快。 结果 10 个 Agent 跑了 15 小时,讨论 733 次,搞出了一个新算法 C-HD。 在一类特定稀疏图上,Dijkstra 是:O(n log n) C-HD 做到了:O(n log^(11/12)n) 看起来只少了一点,但理论算法研究里,这种从已经研究烂的问题里再抠出一点渐近提升,本来就是最难的活之一。 而且Agent们不只给算法,还把证明一起写了。 289 个 Lean 文件,最后直接交给 Lean Kernel 做机器验证。 Lean 已经证明:在他们定义的计算模型和那一小段图密度范围内,C-HD 的确能正确求最短路径,并达到他们声称的复杂度上界。 人类扔进去一个开放研究问题,10 个 Agent 自己试错、互相挑错,15 小时后交回一个新算法 + 可机器验证的证明。 过去要人类研究员几个月甚至几年的试错的研究过程,现在开始可以让Agent并行复制了。
1
6
2,260
When Everyone Wants to Replicate Taiwan: The Semiconductor Supply Chain Enters the Age of Localization tspasemiconductor.substack.c…
1
3
13
1,306