字号·· | 护眼
日经

印度芯片设计初创企业难以跨越商业化的最后一道坎India's chip design startups struggle with final commercial hurdle

点「原文对照」整页切到原文,或双击某段只看那段的原文。

由于资金限制和较长的开发周期,印度的芯片设计初创企业难以实现必要的半导体产量,从而无法在市场上形成有竞争力的价格。© 路透社Due to capital constraints and a longer development cycle, India's chip design startups are struggling to produce semiconductors at the volume necessary to make them competitively priced in the market. © Reuters

班加罗尔 —— 在政府支持的推动下,印度的芯片设计初创企业正逐步接近商业化生产,但因产量低、供应链瓶颈导致的高昂晶圆代工成本,正使其上市之路变得复杂。

BENGALURU -- Propelled by government support, India's chip design startups are nearing commercial production, but steep fabrication costs due to low volumes and supply chain bottlenecks are complicating the path to market.

受限于资金和较长的开发周期,一家刚开始生产的初创企业可能只会从晶圆厂采购约 100 片晶圆,每片晶圆包含几千颗芯片,具体数量视芯片尺寸而定。相比之下,大公司每月下单的晶圆数量达数万片。

Due to capital constraints and a longer development cycle, a startup just beginning production would utilize perhaps 100 wafers from a fabrication plant, each containing a few thousand chips, depending on its size. But in comparison, bigger companies place orders for tens of thousands of wafers a month.

“这些生产晶圆厂,它们看重的是产量,”已为电力、电信等领域的芯片完成“流片”(即设计最终定稿阶段)的 Vervesemi 创始人 Rakesh Malik 表示,“它们向我们收取的价格比给其他公司高出 50%。”

"These production fabs, they look for volume," said Rakesh Malik, founder of Vervesemi, which has reached the final stage of design, known as a tape out, for chips for sectors ranging from power to telecoms. "They charge us 50% more than what they give to other companies.

Malik 补充道:“这是一个成本挑战,因为当你试图为产品定价时,很难具有竞争力,因为你从晶圆厂采购的原材料或晶圆价格不具竞争力。”

"It is a challenge in terms of cost because when you are trying to price your product in the market, it is very hard to compete because your raw material or the wafer, which is coming from the fab, is not priced competitively," Malik added.

尽管如此,仅仅入选印度的半导体计划,就为此前靠自筹资金苦撑多年的初创企业打开了大门。今年,Vervesemi 融得 1000 万美元风险投资。6 月,总部位于班加罗尔的初创企业 Fermionic Design 宣布与美国晶圆代工厂商 GlobalFoundries 达成商业合作,制造射频芯片,成为首家实现此类合作的印度公司。

Still, simply being chosen for India's semiconductor mission has opened doors for startups that had previously spent years bootstrapping. This year, Vervesemi raised $10 million in venture capital. And in June, Bengaluru-based startup Fermionic Design announced a commercial partnership with U.S.-based fab operator GlobalFoundries to manufacture radio frequency chips, becoming the first Indian company to do so.

Fermionic 创始人 Gautam Singh 将这一合作能在一定程度上缓解产量难题归功于半导体政策环境,并表示与 GlobalFoundries 合作后,公司已实现“中等”产量出货,并计划循序渐进地扩大规模。

Fermionic founder Gautam Singh credits the semiconductor policy environment for how the partnership has been able to address the volume challenge to some extent, adding that with GlobalFoundries, the company has been shipping "medium" volumes and plans to scale up slowly.

他说:“从 GlobalFoundries 的角度来看,印度将成为半导体价值链中更重要的一环。建立这种合作表明,他们能在潜在未来客户生产初期就提供支持。”

"From the GlobalFoundries point of view, India is going to be a bigger part of the semiconductor value chain. Having this collaboration signals they can support potential future customers even during the initial stages of production," he said.

本地芯片设计公司的增长被视为一项重要的政策成果,因为创办此类初创企业风险极高。印度拥有全球20%的芯片设计工程师,但他们大多就职于高通和英特尔等跨国公司的本地办事处,这使得想创业的人更难放弃这些备受青睐的工作,去创办一家在该行业可能需要十多年才能建立客户群并实现盈利的公司。

The growth in local chip designers is seen as a significant policy success, as launching such a startup is a risky proposition. India hosts 20% of the world's chip design engineers, but mostly at local offices of multinational companies such as Qualcomm and Intel, making it harder to tempt a would-be entrepreneur away from one of these coveted jobs to start a company in an industry where it can take more than a decade to build a customer base and turn a profit.

在2021年底启动的印度半导体任务第一阶段中,24个芯片设计项目获得了政府资助,覆盖50%的芯片设计成本,并提供高达净销售额6%的部署激励,分别上限为每家公司1.5亿卢比(约158万美元)和3亿卢比。

Under the first phase of India's semiconductor mission, launched at the tail-end of 2021, 24 chip design projects received government funding covering 50% of chip design costs and deployment incentives of up to 6% of net sales turnover, capped at 150 million rupees ($1.58 million) and 300 million rupees per company, respectively.

此外,超过100家公司在班加罗尔的一个中心免费获得了昂贵的工业级电子设计自动化工具使用权,班加罗尔常被称为印度的“硅谷”。

In addition, more than 100 companies were given free access to costly industry-grade electronic design automation tools at a center in Bengaluru, which is often called India's Silicon Valley.

“我们实际上从未想过会获得如此大规模的芯片设计基础设施采用响应,”参与该任务的一位政府官员在匿名条件下表示,并补充说,此前州政府的类似举措并未看到如此大的需求。

"We actually never imagined that we would get such a massive uptake for this chip design infrastructure," a government official who is part of the mission said on condition of anonymity, adding that similar initiatives by state governments previously didn't see as much demand.

在今年启动的半导体任务第二阶段中,政府将向初创企业报销芯片净销售额的9%。不过,这位政府官员承认,大多数初创企业在通往市场的道路上,不可避免地仍将在产量和成本方面苦苦挣扎。

Under the second phase of the semiconductor mission, launched this year, the government will reimburse 9% of the net chip sales for startups. Still, most startups will inevitably struggle with volumes and cost in their path to market, the government official acknowledged.

他补充说,即使塔塔集团的印度首座晶圆厂开始商业化生产,它提供的服务也不太可能与印度初创企业设计的多种芯片相匹配。

He added that even when India's first fab by the Tata conglomerate starts commercial production, it is unlikely that the services it offers will align with the variety of chips Indian startups are designing.

“坦率地说,这是一个自由市场。强制初创企业去塔塔,或要求塔塔给出优惠价格,这在纸面上行得通,但在实践中行不通,”他说。

"Honestly speaking, it's a free market. Forcing a startup to go to Tata or asking Tata to give a favorable price, that will work well on paper, but not in practice," he said.

尽管印度目前已有五家芯片封装厂投入商业生产,但初创企业在定价方面的困境在芯片封装环节依然存在。晶圆代工厂(如台积电)根据芯片设计,将微观电路设计逐层印制在硅晶圆上后,封装厂会将这些完成的晶圆切割成单个芯片,并将其封装在保护外壳中。

The pricing conundrum for startups also persists at the chip packaging stage, despite India now having five such plants in commercial production. After a fab, such as Taiwan Semiconductor Manufacturing Company, prints microscopic circuit designs layer-by-layer onto silicon wafers based on the chip design, a packaging plant cuts those completed wafers into individual chips and seals them in protective casings.

Vervesemi 的马利克表示,印度大多数封装厂给他报出的价格,比国外工厂的报价高出两到三倍。尽管如此,海关方面的困难仍可能迫使芯片设计公司转向这些更昂贵的国内工厂,Vervesemi 便与位于古吉拉特邦的 Suchi Semicon 就其某款设计建立了合作关系。

Vervesemi's Malik said the pricing that was quoted to him from most of India's packaging plants was two to three times higher than what was quoted at factories abroad. Nonetheless, difficulties with customs can push chip designers toward these more expensive domestic plants, with Vervesemi partnering with Gujarat-based Suchi Semicon for one of its designs.

一家初创企业送往国外封装的不仅仅是一颗芯片,而是包含数千颗芯片的晶圆。它收回的也不是晶圆,而是 5000 颗已切割、分离并封装好的芯片。这种差异让海关官员感到困惑,Vervesemi 花了三个月才理清这一沟通误会。

A startup sends not just one chip but the wafer containing thousands of chips outside India's borders to be packaged. What it gets in return is not a wafer but 5,000 sliced, diced and packaged chips. That discrepancy has confounded customs officials, and it took Vervesemi three months to untangle the miscommunication.

另一个类似的障碍是:印度芯片设计公司预先向海外晶圆厂支付费用制作光掩膜版,这是一种高精度版材,作为印制微芯片电路的模板。印度央行规定,通过预付汇款支付的实物资产必须在三年内实物进口到印度,但光掩膜版本应留在晶圆厂内。马利克说:“我们目前也在为此奔走协调。”

In another such hurdle, Indian chip design firms pay overseas fabs upfront to make a photomask, which is a high-precision plate that acts as a stencil for printing microchip circuits. India's central bank rules mandate that a physical asset paid for via advance remittance must be physically imported into India within three years, but masks are meant to stay in the fab. "We are fighting this also right now," Malik said.

这位政府官员将这些小插曲归因于州政府、部委和海关官员对半导体行业的理解仍处于起步阶段,并补充称,当初创企业因监管问题导致货物滞留时,官员们已在不到一天的时间内进行了干预。

The government official attributed these hiccups to the still nascent understanding that state governments, ministries and customs officials have of the semiconductor sector, adding that officials have intervened in less than a day when startups find shipments stuck due to regulatory issues.

他补充说,在半导体任务的第二阶段,优先事项是培育已经生产出芯片的初创企业,而不一定是扩大基础以资助更多公司。但一些行业资深人士主张采取不同的方法。

He added that under phase two of the semiconductor mission, the priority is to nurture the startups that have already produced a chip, rather than necessarily expanding the base to fund more companies. But some industry veterans have advocated for a different approach.

“我认为24个项目太少了,”初创企业Orvis Semi的创始人穆库尔·萨卡尔(Mukul Sarkar)说,他指的是第一阶段资助的项目数量。“就这个国家的规模而言,仅在德里就应该有25到30家公司。

"I think 24 is too small a number," said Mukul Sarkar, founder of startup Orvis Semi, in reference to the number of projects funded under the first phase. "For the scale of this country, 25 to 30 companies should be in Delhi alone.

“资助初创企业的基本经济原则不会改变。在这些公司中,只有两到三家能撑过下一个十年。”

"The fundamental economic principle [of funding startups] will not change. Out of these companies, only two to three will see the next decade."

阅读媒体原文 ↗