How Innovation Died at Intel

[Xebec]

Grove was a legend and a force but I could also say he failed as a founder as he created a leadership and culture that failed in the following decades as was his spectacular failure in succession planning.

One can argue Intel did rise to process leadership. What they did strategically was blah milk the opportunities that presented to them. To be honest during the Grovian days process leadership didn’t exist that came later and was squandered in the end by failed senior management in the decades that followed. Intel while innovative in process leadership was saddled with management myopia and a strategy that was so x86 focused. This was classical Innovators Delma if there ever was

Totally agree on point 1. I think this is similar to how the vast majority of family businesses fail by the 3rd generation - the culture isn't established in a way that can be maintained. (or isn't adaptive enough to changing times). IMO, a high performance culture is challenging (but not impossible) to maintain long term without burning out large amounts of employees.

Re: Process leadership - I'm not sure which is true here. When I look at Intel's competition in the 80s and 90s, Intel was often one step ahead on process. The 68040 (1989-1990 release) couldn't maintain clock speed parity with the 486 which eventually buried it -- largely because of lack of newer processes. AMD's K6 (1996) and K7 (1999) were largely kept at bay by Intel getting to a process ahead of AMD, and Intel was at 130nm well before anyone else (2001 Pentium 3 Tualatin). Intel also had a few interesting interim processes like the Pentium MMX getting a ".28 micron" process, distinct and performant vs. the .35 used on the last of the vanilla Pentium's.

"Tick tock" and "Copy EXACTLY" get the credit for scaling out regular (aggressive) process updates, but it looks like Intel executed pretty well on logic shrink under Grove's tenure (and the first few years after).

 

[krsree]

When you are ahead dosing on your business and strategy innovation. Can be very different. When you are ahead it’s easy to become arrogant especially if you are a westerner. Yeah that is a stereotype. It takes a special upbringing and company culture of many to drive to become a leader and keep the humble hard working and paranoid themes.

People over play Andy’s only the paranoid survive mantra as something that is deep at Intel. There is little to non existence of the kind of paranoia and the scrutiny go of the big and little details at Intel today necessary to survive.

Now you got a bean counter and a sales marketing executive leading the company. I am certain they are top 1% and can and do understand a few things about technology and manufacturing, business and design. Chip making and product design as well as Technology and manufacturing is about the on the ground boots details. Nobody who hasn’t been on the ground and fought those has an appreciation for those. Like a general who has never been on the front lines in a war will simply not have the detailed experience to ask the hard questions that force the organization to focus on the critical little things that win battles and wars. Intel is in need of a transformation and that is big bold strategy but more important detailed scrutiny and execution of details.

Across TSMC the senior leadership has that paranoi, experience, culture and ask the kind if detailed question that force activity and organization to execute on the details.

At Intel the senior leadership across the manufacturing side and TD, all the way up to Pat had none of that, just professional managers and until a new senior leadership comes Intel is destined to fail. Even Pat didn’t possess nor pretended to do those things so in the long run he was going to fail.

Ironical that it was Intel leadership that was once famous for being paranoid!

 

[BruceA]

Totally agree on point 1. I think this is similar to how the vast majority of family businesses fail by the 3rd generation - the culture isn't established in a way that can be maintained. (or isn't adaptive enough to changing times). IMO, a high performance culture is challenging (but not impossible) to maintain long term without burning out large amounts of employees.

Re: Process leadership - I'm not sure which is true here. When I look at Intel's competition in the 80s and 90s, Intel was often one step ahead on process. The 68040 (1989-1990 release) couldn't maintain clock speed parity with the 486 which eventually buried it -- largely because of lack of newer processes. AMD's K6 (1996) and K7 (1999) were largely kept at bay by Intel getting to a process ahead of AMD, and Intel was at 130nm well before anyone else (2001 Pentium 3 Tualatin). Intel also had a few interesting interim processes like the Pentium MMX getting a ".28 micron" process, distinct and performant vs. the .35 used on the last of the vanilla Pentium's.

"Tick tock" and "Copy EXACTLY" get the credit for scaling out regular (aggressive) process updates, but it looks like Intel executed pretty well on logic shrink under Grove's tenure (and the first few years after).

I’d say it wasn’t till maybe P856 0.35um Intel achieved parity. Other had competitive processes but didn’t have a platform or design to milk it to the max. IBM CMOS was superior but didn’t have any volume product to showcase it. The same could be said of the other chip companies at the time.

With the volumes of x86 from 386 and beyond with Intels IDM and binning enabled them to put a process cadence that beat everyone even the well funded alliances from 858 all the way to 1272. It also gave them some terrible or great practices that hamper them today. The two year process cadence with CE and tick tock was a gold mine decades and gave them scal and leadership as well as creates a unique culture. They were so far ahead that to rap the volumes they wanted at the density they desired 1274 they had no option but to do quad pattern and other tricks that simply they underestimated. It was simply not an option to not make an aggressive scale at that time. Their GM had to justify value to his internal partners and that was what in the end broke the company. That same GM never gave or empowered the Foundry group at the time the people or priority to deliver, ironic as at that time their leadership was complete as was their ability to spend money to support the foundry but alas they were foundry pretenders.

LTD was and still is of such arrogance and do no wrong and listen to nobody. They have a factory network all the way to the highest leadership that couldn’t make a technology decision it was CE!, From that CE came they got BK and that was that. Combined with some terrible leaders in LTD and that was that. The seeds were sown well before 10nm but their arrogance and ignoring of the changing landscape and ARM, as well as Foundry is astound failure of both the leaders and board.

TD still has some brilliance but it is saddled with leadership that has no clue to what is going on outside or how to revamp the organization to be competitive

 

[osnium]

I’d say it wasn’t till maybe P856 0.35um Intel achieved parity. Other had competitive processes but didn’t have a platform or design to milk it to the max. IBM CMOS was superior but didn’t have any volume product to showcase it. The same could be said of the other chip companies at the time.

With the volumes of x86 from 386 and beyond with Intels IDM and binning enabled them to put a process cadence that beat everyone even the well funded alliances from 858 all the way to 1272. It also gave them some terrible or great practices that hamper them today. The two year process cadence with CE and tick tock was a gold mine decades and gave them scal and leadership as well as creates a unique culture. They were so far ahead that to rap the volumes they wanted at the density they desired 1274 they had no option but to do quad pattern and other tricks that simply they underestimated. It was simply not an option to not make an aggressive scale at that time. Their GM had to justify value to his internal partners and that was what in the end broke the company. That same GM never gave or empowered the Foundry group at the time the people or priority to deliver, ironic as at that time their leadership was complete as was their ability to spend money to support the foundry but alas they were foundry pretenders.

LTD was and still is of such arrogance and do no wrong and listen to nobody. They have a factory network all the way to the highest leadership that couldn’t make a technology decision it was CE!, From that CE came they got BK and that was that. Combined with some terrible leaders in LTD and that was that. The seeds were sown well before 10nm but their arrogance and ignoring of the changing landscape and ARM, as well as Foundry is astound failure of both the leaders and board.

TD still has some brilliance but it is saddled with leadership that has no clue to what is going on outside or how to revamp the organization to be competitive

Have the technology problems in recent years humbled LTD? I would think all the humiliation would have been humility into them.

 

[nghanayem]

Have the technology problems in recent years humbled LTD? I would think all the humiliation would have been humility into them.

Completely and utterly. Ann K has even publicly talked about adopting industry BKMs, processes, and equipment.

 

[osnium]

Completely and utterly. Ann K has publicly talked about adopting industry BKMs, processes, and equipment.

Thank you for your reply. I was curious because your observation is different from BruceA’s observation.

 

[BruceA]

Completely and utterly. Ann K has publicly talked about adopting industry BKMs, processes, and equipment.

Because she said so, or we have evidence from direct observation and action as and results of such ?

 

[nghanayem]

Because she said so, or we have evidence from direct observation and action as and results of such ?

Public evidence of culture change include various recent papers Intel has shown that include tidbits about various changes to their development model and how they do development vehicles, and crediting these changes with increasing the rate of learning. There is also the public proof point of development not being stalled out anymore. Or 14A being a mobile optimized process out of the gate rather than only being intended for HPC like the nodes that were defined in an "IDM 1.0" world and telling customers something to the effect of: "This is our technology, it is tuned for the requirements of CCG, and we will not meaningfully change anything beyond what we have. So use it or don't.". (I at least expect this is more or less how that sort of conversation would have happened during the intel custom foundry days). You also have Intel being much more conservative on their process technology than they were at "14nm", "10nm", and old "7nm". You can also listen to the GM of MediaTek USA at the last IFS DC keynote for a fairly glowing picture of how they saw Intel foundry evolve during the course of getting their products off the ground:

Sure I think I can kind of put it into three kind of buckets,
if you will. One was the experience around the engagement and enablement side.
And in that particular bucket, as I'll call it, I guess, is, the team really
looked to align with your teams to figure out, how do we optimize? How do we refine the designs?
How do we work best together to maximize power and performance in area. And I think what we found the team was very pleased with,
we got to a point where Intel Foundry really leaned in. In many cases put people on site from an engagement perspective,
working side by side to help us with those things, to help us make sure that that we understood,
how your process worked and it was optimized with the tools and then, your team understanding what our design priorities were.
And so that design engagement was really critical. And that enablement side and there's multiple instances where
the team was really pleased with the engagement and the reaction that Intel took to support us.
The second part of this is just the overall interaction around the fab and fab operations and manufacturing readiness.
The team really felt that Intel listens to feedback and that they were open to that feedback. That they they would take
the time to thoroughly understand the issue and then respond in a way that that addressed the issue.
And it might have taken a little bit of time to understand it. But once we figured it out and there were commitments in place,
Intel quickly executed to it. I would say that that
the comment was made that there is exceptional capability from a fab operations perspective from our team.
The one challenge we had early on was in the test chip yields, as you're well aware.
And the team was really pleased with how quickly the Intel team reacted and worked with our team.
And in less than nine months, we went from really challenging yields to very,
very competitive yields. And the team, the word that they they used was remarkable progress
on that. So the yield ramp was exceptional. You know once we got the challenges fixed.
And then the third part that I would comment on is just the overall mindset in the culture of Intel Foundry, focused
on our success. We believe that that Intel truly understands what it's required to be a foundry partner.
What we need as a customer. We think that they've made significant adjustments along the way, both organizationally and in their process
and workflows over the last couple of years. We found them, like I said earlier, to be very open minded in their approach, very customer oriented
and making us successful. And so overall, the team is very confident that with
the level of collaboration and teamwork we have with the Intel Foundry team and the commitment just to continuous improvement,
that the partnership is going to allow us to really effectively service our customers going forward.

Even with all of their technical prowess, the TMG of the 2000s would have never had a customer describe them thus. Is everything 100% perfect? No. But any learning organization would tell you they have deficiencies and that there is room for improvement. Anyone who says they are perfect is lying to themselves or to you. But clearly they (intel foundry) understand what was being done was not working, and they couldn't just keep flogging their employees to work harder and put in more overtime to human glue a MacGyvered solution together. The results over the past couple of years clearly indicate that the TMG of today is willing and able to learn how to be successful as the manufacturing arm of an IDM, and more importantly for the future and growth of the wider Intel corp, as a foundry.

 

[OneEng]

Yes, Intel made many mistakes over the last 20 years and these have been well documented and discussed to death already. But if you look at those mistakes they are mainly business and strategy failures, not innovation failures.

It has been mentioned in this thread already that AMD also suffered from mis-steps. Ironically, I would argue that AMD did, for sure, but they were financially forced to abandon bad paths much earlier than Intel.

Both AMD and Intel have found themselves in a new reality where new processes are exponentially more expensive while simultaneously providing only incremental improvements over the previous nodes.

TSMC has approached this problem by methodically and conservatively advancing new nodes, charging a premium for these new nodes and maintaining lots of designs that utilize older nodes for a very long time. This gives TSMC LOTS of volume to amortize the cost of the node transition over.

AMD suffered FIRST from this new reality. Sadly for Jerry Sanders, his "Real Men Have Fabs" was realized a day late and tons of dollars short. Intel simply had more volume to spread their costs across and for a while longer, that was enough.

It is my opinion that Intel can not survive with the vertically integrated philosophy.

Yes, the decision to move to EUV was very late. That broke the chain originally and allowed TSMC to surpass them. Still, the bigger trend that killed their profit was the exponentially higher cost to move to a new node and the much more meager improvements to PPA for each new node.

Missing the boat on mobile CPU's, partnering with Apple, etc, etc, yes, those were big blunders as well; however, I still believe the biggest strategic miss they had was not seeing the macro trend I have described.

I have heard many people (Including Pat G) that everything is riding on 18A. I disagree. Intel can become a successful CM, but doing so will take a huge commitment to change. 18A may well be a very good node, and Intel may put out some really great Intel products on it, but that is a far cry from what needs done in order to create all the tools and processes and business modifications needed to onboard different levels of customers and their products to Intel fabs.

IMO, this will take years.

Meanwhile, Intel can't keep burning money at the rate they have been. The 5N4Y strategy was God awful expensive. Changing the entire business as I mentioned above will also be expensive. I see Intel's biggest challenge not in their innovation (which I agree has actually been very good in their history), but in creating a money making foundation again.

 

[Maximus]

Meanwhile, Intel can't keep burning money at the rate they have been. The 5N4Y strategy was God awful expensive. Changing the entire business as I mentioned above will also be expensive. I see Intel's biggest challenge not in their innovation (which I agree has actually been very good in their history), but in creating a money making foundation again.

In Pat's era, the logic flow appears to be:

1. Maintain the foundry: Intel's primary issue is the delay in process nodes, necessitating the 5N4Y strategy to regain leadership.

2. Build numerous fabs: To generate sufficient momentum and offset the substantial investment in process development, a significant number of fabs are required.

3. Address funding shortages:
(a) Seek financial support from other countries' governments (e.g., Germany, Israel).
(b) Obtain funding from the US government (e.g., through the CHIPS Act).
(c) Collaborate with private funds.

4. IDM 2.0 proposal: Due to insufficient chips to supply the new foundry.

Following this foundry/process-centric logic, prioritizing new breakthrough products would not be a high priority.