With Shrink Ending, What's Next?

[Daniel Nenni]

Having worked for both ASIC and small/large system companies, I think you're a bit off the mark -- the biggest cost driver (apart from process) is design complexity, especially software.

I do not include software as part of tape-out. Software can be a big swing depending on the application.

The majority of SoCs I see are variations or have reusable blocks. Now with chiplets and multi die packaging tape out costs can be controlled even more. It really isn't as bad as it seems.

Unfortunately inflated costs are a big problem at the moment but as demand meets supply once again that ship will right itself. 2023 should bring back some normalcy to the semiconductor industry so I see smooth sailing on the horizon, absolutely.

 

[Paul2]

I do not include software as part of tape-out. Software can be a big swing depending on the application.

The majority of SoCs I see are variations or have reusable blocks. Now with chiplets and multi die packaging tape out costs can be controlled even more. It really isn't as bad as it seems.

Unfortunately inflated costs are a big problem at the moment but as demand meets supply once again that ship will right itself. 2023 should bring back some normalcy to the semiconductor industry so I see smooth sailing on the horizon, absolutely.

Normalcy? Nowhere on the horizon.

We have one gigantic global event after another lined up on the horizon.

And we all know how fragile is the semi supply chain.

Every big chip spender has decreased orders, not so much for huge demand drop, but for expectation of cheap money era absolutely certainly ending.

 

[Daniel Nenni]

Normalcy? Nowhere on the horizon.
We have one gigantic global event after another lined up on the horizon.
And we all know how fragile is the semi supply chain.
Every big chip spender has decreased orders, not so much for huge demand drop, but for expectation of cheap money era absolutely certainly ending.

Do you think the semiconductor supply chain is better or worse? My experience is that it is getting better and much stronger than pre pandemic. Many companies that were buying chips are now making them. The foundries are still posting record numbers. EDA, IP, ASIC, business is still strong.

True, the world is a mess but the semiconductor industry gets stronger, absolutely. Name an industry with smarter, more motivated, more diverse people than the semiconductor industry?

 

[Paul2]

Do you think the semiconductor supply chain is better or worse? My experience is that it is getting better and much stronger than pre pandemic. Many companies that were buying chips are now making them. The foundries are still posting record numbers. EDA, IP, ASIC, business is still strong.

True, the world is a mess but the semiconductor industry gets stronger, absolutely. Name an industry with smarter, more motivated, more diverse people than the semiconductor industry?

The planned expansion by 3 leading semi companies before the Russian invasion was in hundreds of billions of dollars over the term of 4-6 years.

It's an unprecedented CAPEX by any measure.

All these plans were made in the era of cheap money, and they were only possible with cheap money. Now we will see rates going up, and all consumption forecasts are past the threshold of plan review.

My advice. Get a new laptop, and a smartphone, it will last a lot.

The current illusion of good times for foundries are due to:

1. 1. Legacy nodes shooting up in price
2. 2. Leading edge being dominated by few super-high-value SKUs

All of this can be gone very fast. Foundries need to diversify on the leading edge + 2-3 nodes to reduce dependency on top clients.

The industry is heading back to early 200X

 

[Daniel Nenni]

The current illusion of good times for foundries are due to:

1. 1. Legacy nodes shooting up in price
2. 2. Leading edge being dominated by few super-high-value SKUs

All of this can be gone very fast.

Also:

Backlog from a down 2019
Work from home surge
Backlog from the chip shortage
Double or triple ordering to build inventory
Stimulus money in the US
Capital gains surge from the stock and housing market

Just to name a few..... Still, the sky is not falling, semiconductors will go back to single digit growth, all is well in the ecosystem.

 

[VCT]

The planned expansion by 3 leading semi companies before the Russian invasion was in hundreds of billions of dollars over the term of 4-6 years.

It's an unprecedented CAPEX by any measure.

All these plans were made in the era of cheap money, and they were only possible with cheap money. Now we will see rates going up, and all consumption forecasts are past the threshold of plan review.

My advice. Get a new laptop, and a smartphone, it will last a lot.

The current illusion of good times for foundries are due to:

1. 1. Legacy nodes shooting up in price
2. 2. Leading edge being dominated by few super-high-value SKUs

All of this can be gone very fast. Foundries need to diversify on the leading edge + 2-3 nodes to reduce dependency on top clients.

The industry is heading back to early 200X

I have different point of view.
Back to early 200x, TSMC build new Fabs without customers' commitment. Now TSMC build new Fabs with prepayment.
Also for 3nm, 5nm, limited competition from TSMC. Very few customers want to take risk on Intel and Samsung for leading nodes.

Back to early 200x, more competition for leading nodes. You know other competitors will eventually launch leading nodes 1-2 years later. But right now, it's hard to guarantee other competitors can catch up leader 1-2 years later. There is a huge risk of not being able to deliver on time. If it's not for 3nm, it could be 2nm. It's getting harder and harder and the risk for failing is the much higher than 10-20 years ago.

 

[Daniel Nenni]

I have different point of view.
Back to early 200x, TSMC build new Fabs without customers' commitment. Now TSMC build new Fabs with prepayment.
Also for 3nm, 5nm, limited competition from TSMC. Very few customers want to take risk on Intel and Samsung for leading nodes.

Back to early 200x, more competition for leading nodes. You know other competitors will eventually launch leading nodes 1-2 years later. But right now, it's hard to guarantee other competitors can catch up leader 1-2 years later. There is a huge risk of not being able to deliver on time. If it's not for 3nm, it could be 2nm. It's getting harder and harder and the risk for failing is the much higher than 10-20 years ago.

Let's not forget, back in the early 2000s designs (28nm and above) done at TSMC could then be moved to SMIC, UMC, and Chartered. That all changed with the FinFET era, there is no moving designs, so now TSMC can get more favorable terms from customers.