Chinese semiconductor industry

Status
Not open for further replies.

tphuang

Lieutenant General
Staff member
Super Moderator
VIP Professional
Registered Member
Haven't you say don't quote Paul Triolo in this thread?
That's because the discussions around what he was saying got so overly toxic at the time of October sanctions. I have found that he has certain commentaries based on unknown sources that we don't have access to.

For example, he got interviewed here
Please, Log in or Register to view URLs content!
and it seems like we should expect additional Huawei like restrictions on baidu, Alibaba and tencent. This shouldn't surprise everyone here. Any major successful Chinese tech companies are likely to face certain restrictions.

He likely got wind of this from people in DC.
 

theorlonator

Junior Member
Registered Member
That's because the discussions around what he was saying got so overly toxic at the time of October sanctions. I have found that he has certain commentaries based on unknown sources that we don't have access to.

For example, he got interviewed here
Please, Log in or Register to view URLs content!
and it seems like we should expect additional Huawei like restrictions on baidu, Alibaba and tencent. This shouldn't surprise everyone here. Any major successful Chinese tech companies are likely to face certain restrictions.

He likely got wind of this from people in DC.
Given that those companies have lots of American investors and customers, I'd imagine that would p*ss off a lot of people here.

Oh well
 

FairAndUnbiased

Brigadier
Registered Member
That's because the discussions around what he was saying got so overly toxic at the time of October sanctions. I have found that he has certain commentaries based on unknown sources that we don't have access to.

For example, he got interviewed here
Please, Log in or Register to view URLs content!
and it seems like we should expect additional Huawei like restrictions on baidu, Alibaba and tencent. This shouldn't surprise everyone here. Any major successful Chinese tech companies are likely to face certain restrictions.

He likely got wind of this from people in DC.
The best way to build Chinese semiconductor industry is to give the customers no choice but to do so. A fab doesn't want to take risks on domestic equipment because of cost? Alibaba or something can just pay them to do so and give them a guaranteed order of X SoCs.
 

antiterror13

Brigadier
BYD semiconductor's Yangzhou fab has started production. If you want to know the salaries of Chinese fab workers. The attached photo shows that. As you can imagine, costs in China is really low

But perhaps include free accommodation (dorm) and also free meals (?) .. so if you think about it, not really that small. If it is the case, it means that the salary is basically can be saved near 100% ... try doing it in any city in the US or EU
 

tphuang

Lieutenant General
Staff member
Super Moderator
VIP Professional
Registered Member
华为Chiplet专家交流纪要

Q:从中国半导体产业来看,Chiplet技术是否是短期内破局先进制程限制的唯一路线?还有没有其他可能的突破点?

A:目前来看是最有机会的,但是唯一的说法有点太过绝对。Chiplet只是把SOC转换过来,拆分成一个个功能单元再封装起来,其实用package也可以做到,但是会损失尺寸和功率,总体来说Chiplet在PPA上面会更有优势。

Q:中国企业Chiplet技术发展现状,哪些方面领先或者够用了,哪些方面还有不足,是否有部分环节还是有卡脖子问题?

A:要从Chiplet的全产业链来看,Chiplet虽然现在是属于先进封装这一块,但是实际上要涉及到全产业链的联动,包括芯片的设计,EDA,IP,Interposer,材料和设备等,相当于换了一个赛道有更多的机会,但是整体上也和先进制程维持着一样的水平,在芯片制造的后道,设备环节机会好一点,封装这一块是最有机会的。

Q:封测厂方面,和哪家合作比较多?

A:和长电合作比较多,和通富停留在打样试制的程度。

Q:目前产品做到什么程度,涉及到的环节比如EDA,IP与制造有哪些是国产的?

A:国内也有一些EDA公司,国外的EDA软件也能用。已经发布了14nm堆叠得到7nm的服务器的云计算的处理器。EDA部分还是用的原来的国外的EDA,还有部分的国内EDA,华大九天的EDA在试用,IP还是用的国外的,制造就是
Please, Log in or Register to view URLs content!


Q:Die测试量的增加是否能明显带动测试设备商的业绩提升?国内主要是和哪家测试机厂商在合作?

A:测试方面在国外的限制之前还是能测的,国内在扶植华峰和长川,两家给的机会基本是均等的,两家各有侧重点,长川测试的分选会更好一点,测试机更国外差距还比较大。华峰测试机这一块可能稍微更好。

Q:测试方面跟长川的绑定是不是更深一些?

A:长川跟华为的合作意愿更强,希望借助大公司平台把经营方面做好,华峰出于自身考虑,不想绑定在某一家公司或者Chiplet身上,想多元化发展去更好适应市场。

Q:测试增量定量增量有多少?

A:比较难定量测算,举例来说5个功能集成到同一个SOC上,测一次,但是测试比较复杂,耗时比较长,Chiplet需要测五次,但是时间不是简单1:5,由于单个小芯片测试简单了,时间可能变成1:4。另一方面,集成SOC有可能只测关键参数,测一个结果即可,但是Chiplet需要每个小芯片都测清楚,参数不同,需要的时间有可能又会更多,时间比又或许会成为1:6,因此要看具体的产品来定。

Q:Chiplet产品策略怎样?面向鲲鹏的Chiplet,基站Chiplet,还是手机的Chiplet?

A:产品研发从简单到复杂,一般先做大数据处理,云计算方面的需求,就是服务器芯片,成熟以后,再转向消费电子,手机处理器上面去做,什么时候去做要根据具体情况。服务器对体积和散热需求不如手机那么高,手机对尺寸和散热容忍程度是需要解决的难点。

Q:通信基站是不是不需要用Chiplet技术做的14nm?

A:基站要求没有那么高,不一定要用到5nm或者3nm,暂时不需要用到Chiplet,未来也不排除在基站方向用Chiplet,目前来看是没有这方面的推进。

Q:Chiplet技术路线上还有哪些环节可以被美国卡脖子?

A:从产业链来看还是会有卡脖子的地方,单从Chiplet去看,可以自己组合各种功能,但是从芯片制造方向来看还是会有卡脖子的地方,比如14nm芯片制造,还有EDA软件等。此外,虽然封装差距比较小,但是也有关键部件也是能够被卡的。最大的问题是材料这一块,材料的国内外差距也很大,跟设备的差距差不多,现在都是日韩等国,如果美国提升限制,很多材料也没法用,尤其是一些高端基板,低端基板满足不了Chiplet需求,美国卡了基本的话就用不了。先进的基板材料都是被日本垄断,线宽间距和掩膜(ABF膜)也是日本独家垄断,国内自己基板和国外差异还是很大。

Q:
Please, Log in or Register to view URLs content!
也在扩ABF的基板?兴森和深南扩出来的ABF基板能达到要求吗?

A:掩膜是覆盖在ABF上面的材料,这块也有比较大的受限的风险。兴森和深南扩产做ABF基板,但是ABF材料一旦不供应,ABF基本就做不了,线宽线间距也就做不小了。

Q:Interposer的作用,市场空间,分类,单价,用量等?

A:基板占大头30%,接下来就是Interrposer20%,合起来在50%以上,价值量有几十到几百块钱,这是不算芯片的价钱,是除了芯片之外所有材料的价值量。转接板大都是用的硅基的转接板,就是只有晶圆厂能做。

Q:
Please, Log in or Register to view URLs content!
收购的苏州科阳能做TSV,是不是也能做Chiplet?

A:大港目前不做Chiplet,做一些Bumping,但是要看尺寸的。要做到像Chiplet的需求的话应该还做不了,Bumping和rdl是有能力的,但是尺寸达不到chiplet要求,也没有对应的晶圆的后处理的设备。总体来说是能做跟Chiplet相关的一些Bumping,TSV一定需要晶圆厂的设备。

Q:TSV每平米的单价?

A:属于晶圆厂定价,单应该不是按尺寸大小,应该是按加工难度,深宽比来收费。

Q:海外的
Please, Log in or Register to view URLs content!
、肖特,国内的
Please, Log in or Register to view URLs content!
,发布的玻璃基TGV技术,也可以用于interposer,如何评价该技术路线?

A:TGV适用于特殊环境特殊需求,一种是基底做金属化图形,另外一种是做转接板用但是比chiplet要求要简单,间距等比较粗,用非批量性的产品,主要是对热应力变形要求高的特殊领域。起源不是为了用于chiplet,现在要开始改进原有技术路线,适应chiplet的需求,多层、高密度等。

Q:TGV技术过去没有发展起来的主要瓶颈是什么?未来是否有望放量?市场空间多大?

A:玻璃基有难点是通孔定向蚀刻,硅基很成熟,深宽比、保护侧壁等。TSV先刻蚀一个深度然后长钝化层保护起来,然后底部打开保护继续向下蚀刻,玻璃基还做不到。如果这个问题能解决,TGV是有优势能替代TSV的,性能更好,电性能抗磁电感都要好于硅,硅的晶体结构决定了一些缺陷。玻璃更适合短距离快速传递。然后要看TGV和TSV的成本,如果成本一致的话TGV性能有优势,可以替代TSV的市场。

Q:如何评价
Please, Log in or Register to view URLs content!
在TGV、玻璃基封装载板方面的技术实力?

A:沃格最初研发是为了解决光领域的光通信需求有合作,然后看到了chiplet的需求,就想来拓展这块。目前还是光器件领域的TGV能力。另外他还有玻璃基封装产品,目前还满足不了chiplet的需求,和芯片接触面还达不到10+nm级别水平。和华为有技术合作。

Q:Chiplet是否能带动IP厂商(如
Please, Log in or Register to view URLs content!
)的产品虚拟转向实体,IP厂商在Chiplet方向上的弹性有多大?还是说纯IP厂商会失去原有优势?

A:通过IP去找市场,有明确的需求后才会去做,库存不是很大的问题,产品由虚拟转向实体也会带来价值量的提升,此外未来Chiplet的硬IP的组合会更灵活有更多的选择性,因此总体上来说IP厂商可以获取更多的机会

Q:传统EDA在Chiplet项目设计上有诸多问题待解决,对多芯片堆叠的2.5D/3D异构封装显得很低效,芯片和封装相互之间的影响变得更显著,还有芯片可测性设计如何实现均要求DFT工具能够用高度自动化流程实现,国产EDA在Chiplet领域能否追赶上国际龙头的步伐?

A:在Chiplet设计上的EDA相当于换了赛道,的确是对国内EDA厂商追赶国外龙头企业有些帮助,但是海外的龙头发展这么多年还是在细节上优势明显。国内的华大九天有整体架构能力,但是细节还是要解决问题。

Q:Chiplet相关的很多先进技术依然由
Please, Log in or Register to view URLs content!
等把持,因为Chiplet涉及的很多先进封装技术和晶圆级的处理紧密相关,TSV会不会被卡脖子?

A:虽然TSV也是涉及到纳米级别的工艺,但是和SOC芯片制造工艺也不完全一样,要求要等很多,因此可以绕过,不太容易在这个环节被卡脖子。

Q:Chiplet从产业上来讲未来的发展趋势是否真正是高端芯片产品绕不过的技术路线?

A:是目前来说最有优势的一种技术路径,不能说是完全绕不过,未来也许会有更新的颠覆性的技术出现,但是当下来说Chiplet确实具有很多价值

Q:国内《小芯片接口总线技术要求》标准和国际UCIe联盟的协议标准在技术层面有什么本质区别,如果未来UCIe联盟最中国企业进行制裁,是否会造成国内Chiplet技术与国际脱节?

A:国际标准即使封锁了也可以拿过来用,这是不影响的。最差情况只是不带中国厂商制定标准。目前两套标准不会脱节,国内的Chiplet标准发布更多出于是一种制衡的策略
Please, Log in or Register to view URLs content!

Supposedly an interview with Chiplet expert from Huawei. Not sure about its authenticity.
Talk about using domestic vs foreign EDA. You can stack 14nm to get 7nm chip performance. Empyrean's EDA is still getting trialed. Sounds like SMIC is the manufacturer here and they are still using foreign EDA.

In terms of which type of chiplets/stacking to design firm, it says big data center and cloud CPUs first -> smartphone later. Since heat and space requirements for server CPUs aren't as strict as smartphones.

Does not rule out using chipsets for base station, but currently they don't need 14nm chips.
 
Last edited:

FairAndUnbiased

Brigadier
Registered Member
Please, Log in or Register to view URLs content!

Supposedly an interview with Chiplet expert from Huawei. Not sure about its authenticity.
Talk about using domestic vs foreign EDA. You can stack 14nm to get 7nm chip performance. Empyrean's EDA is still getting trialed. Sounds like SMIC is the manufacturer here and they are still using foreign EDA.

In terms of which type of chiplets/stacking to design firm, it says big data center and cloud CPUs first -> smartphone later. Since heat and space requirements for server CPUs aren't as strict as smartphones.

Does not rule out using chipsets for base station, but currently they don't need 14nm chips.
Good interview. He talks about new packaging techniques for chiplet, what companies care about Huawei and which ones don't.
 

HighGround

Junior Member
Registered Member
Yes,I would disagree on this as well. IC is far from the only "remaining shortfall" in China, there are tons of areas where China lags behind and rely on the West to a large degree
This seems to be counter to what I typically hear in this thread. Most people seem to be very optimistic, but I don't share that optimism. Though, to be fair, Chinese progress is a lot further along than what most Westerners think it is.

Are we really going to see a mass production immersion lithography machine in 2024-2025 from SMEE?
 
Status
Not open for further replies.
Top