Gallium Oxide Breaks the Ceiling: When the Fourth-Generation Semiconductor Steps from Lab to 10 kV

氧化镓破顶：当第四代半导体从实验室跨进万伏之门

Ⅰ. A Stone That Redefines Every Volt

第一节 一块重新定义"每伏特"的石头

Silicon ruled power electronics for seventy years—until it hit its physical wall. SiC and GaN pushed the wall further. Now, a transparent ceramic-like crystal with the formula Ga₂O₃​ is walking straight through that wall.

硅统治功率电子七十年，直到撞上物理天花板。SiC 和 GaN 把天花板又顶高了一截。而现在，一种透明、陶瓷质感的晶体——氧化镓 Ga₂O₃——正从墙里穿过去。

Its numbers sound like science fiction:

它的参数读起来像科幻：

• Bandgap ~4.8–4.9 eV​ (silicon: 1.1, SiC: 3.3, GaN: 3.4) — 超宽禁带

• Theoretical breakdown field ~8 MV/cm​ — 3× SiC, ~20× Si

• Baliga Figure of Merit ~10× SiC, ~3000× Si​

• Grown by melt-based methods​ (VB / EFG), same family as sapphire — meaning large wafers without the insane cost of SiC's sublimation growth​

This is why "fourth-generation semiconductor" is no longer a conference buzzword. In March 2026, China's 15th Five-Year Plan​ explicitly named gallium oxide and diamond​ as frontier new materials for industrialization. The race is no longer about who publishes more papers—it is about who ships the first 10 kV die.

这也是为什么"第四代半导体"不再是会议黑话。2026 年 3 月发布的"十五五"规划纲要明确把氧化镓、金刚石列为超宽禁带半导体产业化重点。竞赛不再是论文数，而是谁先交出第一颗万伏级芯片。

Ⅱ. The 2025–2026 Spring Offensive: China Jumps Two Sizes Ahead

第二节 2025–2026 春汛：中国一次跳过两个尺寸

Material: 4″ → 6″ → 8″ → 12″ in 14 months

材料端：14 个月走完 4→6→8→12 寸

The 12-inch jump is not cosmetic. Wafer real estate per run rises ~2.25× vs 8″, and because Ga₂O₃ uses melt growth (not SiC's 2300 °C sublimation in pure Ar), iridium crucible usage can drop ~80%​ with casting/VB optimization. The cost curve bends the right way.

12 寸不是面子工程。单片可用面积较 8 寸提升约 2.25 倍；且氧化镓用熔体法（而非 SiC 那种 2300℃ 升华法），铸造/VB 优化后铱坩埚用量可降约 80%。成本曲线朝正确方向弯了。

Japan's NCT (Novel Crystal Technology) is shipping 6″ (150 mm) β-Ga₂O₃ samples in 2026, targeting 8″ by 2035. China's bulk+crystal+eptx stack is ~1–3 years ahead on wafer size, though Japan still leads on device-level maturity and roadmap discipline (2029 mass production).

日本 NCT 2026 年交付 6 寸 β-Ga₂O₃ 样品，目标 2035 上 8 寸。中国在"晶体尺寸"上领先约 1–3 年，但日本在器件成熟度和路线图纪律（2029 量产）上仍占优。

Device: 4 kV → 9.02 kV → 10 kV+

器件端：4 千伏 → 9.02 千伏 → 万伏级

• Jiufengshan Lab 九峰山实验室 (2026.03): lateral Ga₂O₃ MOSFET on domestic homoepitaxy, 9.02 kV breakdown​ (previous public ceiling <4 kV), double-layer source field plate, no p-type oxide​ — bypassing the p-type stability trap entirely.

• Pinghu Lab 平湖实验室 (Shenzhen, 2026): Ga₂O₃ photoconductive switch >10,000 V, dynamic R_on <10 Ω, sub-ns response, conversion efficiency >80% — enters the 10 kV practical tier.

• Xidian (Hao Yue team): 8 kV heterojunction SBD; 2.77 kV trench SBD (BFOM 1 GW/cm²); Cu₂O/Ga₂O₃ heterojunction 0.83 V turn-on / 2345 V block.

The 9.02 kV result matters not because of the number, but because it was done on Chinese homoepitaxy + Chinese bulk + Chinese device design. No imported p-type, no imported substrate. The chain closed.

9.02 kV 的意义不在数字，而在它是国产同质外延 + 国产单晶 + 国产器件结构闭环出来的。没有进口 p 型，没有进口衬底。链条合上了。

Ⅲ. The Two Old Wounds: Thermal and P-Type

第三节 两道老伤口：散热与 P 型

Every Ga₂O₃ skeptic knows the rebuttal: "Great, but it conducts heat like a thermos and you can't make p-type." 2025–2026 answered both.

每个氧化镓怀疑派都背得出那句反驳："好是好，但导热像保温瓶，而且做不出 p 型。" 2025–2026 把两道都回了。

Thermal: Hao Yue's team inserted a graphene buffer layer between Ga₂O₃ and diamond, pushing thermal boundary resistance to 2.82 m²·K/GW​ — an order-of-magnitude drop. Diamond-on-Ga₂O₃ becomes a packaging architecture, not a dream.

散热：郝跃团队在 Ga₂O₃/金刚石之间插石墨烯缓冲层，热边界电阻压到 2.82 m²·K/GW——降一个数量级。金刚石贴氧化镓从"幻想"变成"封装方案"。

P-type: Jiufengshan's 9.02 kV MOSFET simply didn't use p-type. Double field plates + homoepitaxy modulated the electric field well enough to hit 9 kV without ever needing a p-channel. The "weakness" was routed around, not engineered away.

P 型：九峰山 9.02 kV MOSFET 根本没用 p 型。双层场板 + 同质外延把电场调制好，不到 p 沟道也冲上 9 千伏。弱点被绕开了，而不是被消灭了。

This is the mature move. SiC spent 20 years fighting its own substrate defects; Ga₂O₃ is borrowing that lesson early.

这才是成熟打法。SiC 花了 20 年跟自己的衬底缺陷打架；氧化镓早早借了这堂课。

Ⅳ. Why 10 kV Changes the Game (Not Just the Spec Sheet)

第四节 为什么万伏级改变的是牌局，不是参数表

Power semiconductors live or die by three variables: breakdown voltage, conduction loss, footprint.

功率半导体的命根子是三个变量：耐压、导通损耗、体积。

At 1200 V​ (EV fast-charge, 800 V platforms): Ga₂O₃ competes with SiC on loss, wins on wafer cost and drift-layer thinness​ — same BV, ~1/3 the epi thickness.

1200V（超充、800V 平台）：氧化镓和 SiC 拼损耗，赢在晶圆成本与漂移区厚度——同耐压下外延层薄约 1/3。

At 3300–6500 V​ (rail traction, grid breakers): SiC gets expensive and thick; Ga₂O₃ stays thin, stays cheap, stays cool enough with diamond backing.

3300–6500V（高铁牵引、电网断路器）：SiC 又贵又厚；氧化镓依然薄、依然便宜，加金刚石背板够凉。

At 8000 V+​ (UHVDC, AI data-center busbars, pulse power): this is Ga₂O₃'s own air. Si and SiC need cabinets; Ga₂O₃ needs a fingernail. The 9.02 kV MOSFET and 10 kV photoconductive switch are the first proof that the "thin-dam" promise holds at scale.

8000V+（特高压直流、AI 数据中心母线、脉冲功率）：这是氧化镓自己的空气。硅和 SiC 要机柜，氧化镓只要指甲盖。9.02kV MOSFET 与万伏光导开关，是"薄坝"承诺在尺度上成立的第一份证词。

A 2 m² SiC cabinet shrinks to a chip. That is not an upgrade. That is a category break.

2 平方米的 SiC 机柜缩成一颗芯片。这不是升级，是品类断裂。

Ⅴ. The Geopolitics of Gallium: 95% of the Spigot

第五节 镓地缘：95% 的水龙头

None of this is substrate-neutral. China controls ~95% of global primary gallium supply, and since 2023 has export-controlled Ga/Ge. Ga₂O₃ doesn't need much Ga per wafer—but at national scale, the spigot matters.

这些都不是衬底中立的。中国掌握全球约 95% 原生镓供应，2023 年起对镓锗实施出口管制。氧化镓每片用镓不多——但放到国家战略尺度，水龙头是谁的，很重要。

Japan's NCT can design the cleanest 6″ roadmap in the world; it still sources metal upstream in a market where Beijing sets the valve. The 2025–2026 Chinese burst (Fujia 12″, Garen 8″ epi, Jiufengshan 9 kV) is therefore not just engineering—it is resource + lab + policy​ arriving in the same quarter. The 15th Five-Year Plan naming Ga₂O₃ is the policy half closing the loop.

日本 NCT 可以画出全球最干净的 6 寸路线图；但它的金属上游仍在那个由北京拧阀门的市场里。2025–2026 中国这一波（富加 12 寸、镓仁 8 寸外延、九峰山 9 千伏）就不只是工程——是资源+实验室+政策在同一个季度汇流。"十五五"点名氧化镓，是政策那半边把环扣上。

Ⅵ. The Honest Caveats

第六节 老实人的保留项

• Vertical devices still lag: most record BV numbers are lateral MOSFETs (easy to test, hard to package). Real power modules need vertical Schottky + MOS, where threading dislocation density and ohmic contact on (010) faces remain open.

  纵向器件仍落后：破纪录的多是横向 MOSFET（好测不好封装）。真正功率模块要纵向 SBD+MOS，而 (010) 面位错密度和欧姆接触还没闭环。

• Reliability data thin: 150 °C zero-degradation diodes exist (SINANO, 1148 V), but 10 kV switches need 1000 h HTRB, cosmic-ray ruggedness, and 20-year fleet data no one has yet.

  可靠性数据薄：苏州纳米所有 150℃ 零衰减 1148V 二极管，但万伏开关要 1000h HTRB、宇宙射线鲁棒性、20 年车队数据，都还没有。

• Japan's discipline vs China's velocity: NCT's 2029 mass-production date is credible; Chinese pilot lines are faster but less dated. Both are pre-commercial.

  日本纪律 vs 中国速度：NCT 的 2029 量产可信；中国试点线更快但更少里程。两边都还没商用。

• Diamond integration not yet a product: graphene-interlayered bonding is lab-grade.

  金刚石集成尚未成品：石墨烯缓冲键合还是实验室级。

None of these kill the thesis. They just set the clock: 2026 is the year Ga₂O₃ proved the physics; 2029–2032 is the year it proves the factory.

这些都不推翻命题，只校准时钟：2026 是氧化镓证明物理的年份；2029–2032 是它证明工厂的年份。

Ⅶ. Closing: The Same Loop We Called "Great Medicine"

第七节 收尾：和前文"大药"是同一个环

Trace back to the earlier essays in this thread—the great medicine is immense love and compassion, not fear; the alchemical axis is 甲子乾轴—坎离毂轮; Li Chunfeng's "悟得循环真谛在" reads the outer calendar, Wei Boyang reads the inner one. The same structure shows up in Ga₂O₃:

回看本线程前文——大药是巨大热爱与慈悲而非恐惧；炼丹之轴是甲子乾轴—坎离毂轮；李淳风"悟得循环真谛在"读外历，魏伯阳读内历。同一个结构在氧化镓里又出现一次：

• Fear-driven tech​ races to beat Japan, hoards patents, burns talent → leaky furnace, like anxiety-driven cultivation.

  恐惧驱动的技术急着赢日本、囤专利、烧人才 → 漏炉，正如焦虑驱动的修行。

• Love-driven tech​ sees the 10 kV switch as a way to cut UHV transmission loss so a province drinks more light per ton of coal → the same "immense yes" we called great medicine.

  热爱驱动的技术把万伏开关看成"让一省每顿煤多换几度电"的路 → 正是我们称为大药的那个"巨大的是"。

• The cycle​ is real: 顺行 silicon → SiC → Ga₂O₃ is "inverse alchemy" only if you force it; done rightly, it is 逆行—returning from dispersion (big cabinets, big loss) to concentration (one chip, one thin dam).

  循环是真的：硅→SiC→Ga₂O₃ 若硬推是"顺行"；做对了，就是逆行——从分散（大机柜、大损耗）回聚（一芯、一薄坝）。

Gallium oxide will not "replace SiC" any more than compassion replaces discipline. It opens the 8000 V+ drawer​ SiC couldn't cheaply open, and it does so on Chinese bulk, Chinese epi, Chinese 9 kV MOSFET, Chinese 10 kV switch, written into the 15th Five-Year Plan as a national spine.

氧化镓不会"取代 SiC"，就像慈悲不取代戒律。它打开的是 8000V+ 那只 SiC 没法便宜打开的抽屉，而且它是踩着中国单晶、中国外延、中国 9 千伏 MOSFET、中国万伏开关、写进十五五当国家脊梁的方式打开的。

The stone is plain. The voltage is not.

石头很平常。电压不平常。

继续阅读：氧化镓   第四代半导体

此文由 怡心湖 编辑，若您觉得有益，欢迎分享转发！：首页 > 常识论 » 氧化镓破顶：当第四代半导体从实验室跨进万伏之门