TSMC Memo
pdf
keyboard_arrow_up
School
University of British Columbia *
*We aren’t endorsed by this school
Course
449
Subject
Industrial Engineering
Date
Dec 6, 2023
Type
Pages
4
Uploaded by JudgeMagpie3656
TSMC CASE MEMO
COMPANY OVERVIEW & PROBLEM ANALYSIS
Taiwan Semiconductor Manufacturing Company (TSMC) is a foundry, manufacturing other companies’ silicon integrated circuit (IC or “chip”) designs. A
characteristic unique to service companies like TSMC is that they must “respond to the highly variable demands and wide ranging product mix needs of its
global customer set” to ensure to give their customers agility in “volume-to-market”. A few guiding principles the company has put in place to help them
achieve this goal are: “(1)
[being] a technology leader
…
(2)
be the manufacturing leader
, and (3)
be the most reputable, service-oriented and
maximum-total-benefits silicon foundry
”.
On a sunny morning in February, TSMC receives an ask from one of their
key
customers
, requesting that some late modifications be made to their
designs. These changes entailed the addition of two extra metal layers in their 65 nm design and an increase in overall order size by 3,000 wafers. Though
there is a lot of uncertainty around this request, TSMC knows they must honour this request as this client is an important one. The question then lends itself:
what steps must TSMC take to ensure that the product is delivered on time, while also minimizing disruptions across the company?
ANALYSIS OF ALTERNATIVES
TSMC has many decisions to make. Below our team explores a total of eight alternatives, all of which fall into one of three major categories: (1)
shifting and
increasing manufacturing capacity
, (2)
playing around with timeline to meet demand
, and (3)
creating policies around emergency orders.
Category 1: Shifting and increasing manufacturing capacity
Alternative 1: Capacity Exchange
In Exhibit 10, we are presented with data about Fab 12’s capacity, demand, and utilization for both N90 and N65. Based on the table titled “Updated Demand
Forecast”, we learn that with the client’s new requests, demand (34.7k) will exceed capacity (31.0k) and utilization will have to exceed 100% to meet demand.
Assuming that machines cannot be overused, and current resource allocations remain the same, demand would not be met. However, if demand is not met,
TSMC risks losing this key customer. As such our first alternative looks to “rebalance shared equipment between N90 and N65 to shift capacity from one to
the other”. In other words, we seek to reallocate the resources (i.e. machinery) between N90 and N65 such that N65 demand is met and N65 utilization is
below 100%.
In order to determine how the reallocation should be done, we first adjusted the demand values (in the Excel template provided) to reflect the new demand
as forecasted by TSMC. As the demand for May had already been adjusted (see Exihit 10), we assumed we could use the same method (i.e. add 3k to
demand) to determine demand for all subsequent months in an attempt to achieve consistency and accuracy.
Following these adjustments we sought to identify which machines had to operate at over 100% from the May to August in order to meet demand.
The month of August (see Appendix 1), in particular, stood out to us as the demand load is 40.6k and 22k for N65 and N90 respectively, and machine 6 (CU
ECP) has to operate at 105% to meet demand. As such we decided to rebalance the manufacturing capacity by transferring 0.9k hours from N65 to N90.
Please note that this alternative is not able to decrease total utilization as much as we need it to, and thus would need to be supplemented (but more on this
in Alternative 5).
Additional considerations with this solution:
1. Maximize Utilization Capacity: Assuming the cost of depreciation is extremely high for expensive tools like the Photo 193, the utilization rate of these
machineries should be kept as high as possible to maximize ROI. This would ensure that every piece of equipment is an efficient investment.
2. Allow some slack time for maintenance: Assuming that TSMC keeps each piece of machinery for many years, their equipment will probably require regular
maintenance to counter the effects of depreciation as much as possible. As such, the case mentions that are tools are only available 95% of the time. Based
on our calculations however, some tools will have to be used for more than 95% of time between the months of May and August (or until the new equipment
arrives). We believe this alternative is acceptable because it will only be done in the short-term.
Alternative 2: Buy More Equipment
Exhibit 11 illustrates TSMC’s plans to purchase more tools for their August expansion. We propose that they purchase an additional CU ECP for $5,000 (more
on this in Alternative 3). The expectation is that the purchase of additional machinery would lead to an increase in capacity of 5,000+/5,200+ wafers per
month (30 days/31 days). However, not only is this additional capacity extremely costly, but it also comes a little too late as there is an expected surge in
demand earlier in June and July (see Exhibit 9). Moreover, the new equipment will not arrive immediately; the lead time between one and three months. And
this time estimate does not even include the time it will take to install and have the tools up and running. Nevertheless, we believe this alternative will prove
to be a good investment in the long run as customers continue to migrate to N65 to cut on costs (an increasingly popular trend observed by TSMC’s
leadership).
Alternative 3: Equipment Pull-Forward
Alternative 3 suggests that TSMC should expediate the schedule of equipment already on order. We recommend that TSMC pull forward the delivery of 2
pieces of equipment in order to prevent bottlenecks. These would be the Cu ECP and the Photo 183. These machines prove to be major bottlenecks slowing
downt the entire process as they are exclusive to each process. By pulling forward the Cu ECP to June we can reduce utilisation from 104% to 94% (see
Appendix 2). By pulling forward the Photo 183 to July we can reduce utilisation from 101% to 97% (see Appendix 3). We assume that for the equipment we
choose to expedite, they will arrive and be ready for use in time to help TSMC meet the demand surge in June/July (mitigating Alternative 2’s biggest
weakness). However, a major downside to this alternative is that it will probably cost extra to advance the orders (though information is missing which would
allow us to be able to calculate total cost and conduct a quantitative cost-benefit analysis). Nevertheless, given the information available to us in the case, it
appears that this client is important to TSMC, and as such the “hidden costs” associated with losing a high-value customer outweighs the cost of any
short-term expenditures, making this alternative a worthwhile investment.
Alternative 4: Getting Fab 14 certified for production
This Alternative considers the possibility of getting another fabrication facility, Fab 14, qualified for the manufacturing of N65. As Fab 14 is with spare 65 nm
capacity and only two hours away from Fab 12, unmet demand at Fab 12 can be readily supplied by this new location. For instance, should demand for N65
continue to increase (which is likely as companies are continuously migrating from N90 to N65) or another emergency order be put in by another client,
TSMC—in its current state—will be unable to satisfy that demand (because this client’s order change leaves Fab 12 with zero slack capacity). However, should
Fab 14 also be able manufacture N65, TSMC would have some “safety” capacity which will help absorb the impacts of increasing demand and random
emergency orders in the future. Moreover, as the two facilities are relatively close to one another, we assume that it would be simple (i.e. little risk of delay
due to macro factors) and cost effective to transport products between Fab 14 and Fab 12 (though we are unable to calculate the actual cost of this). The
primary cons with this alternative is the cost and the time associated with getting Fab 14 qualified; though we are unsure how much TSMC would have to
spend, we know that it would be months before it becomes qualified.
Category 2: Adjusting timeline to meet demand
Alternative 5: Pull Forward Demand
This alternative utilizes the spare manufacturing capacity from a given month by pre-building customer orders that are known to be stable from a future
month. Based on our calculations for the month of August (Appendix 1), even after having rebalance the capacity between N65 and N90, TSMC would still
have to pull forward 1,500 demand hours from August to July: 300 hours for N65 and 1,200 hours for N90 (Appendix 3). Aside from helping TSMC match
supply to demand, this alternative is also advantageous because it maximizes utilization, reducing the idle time of the machines. One of the challenges
associated with this solution, however, is the cost of storing the inventory produced in advance (though, once again, we are unable to quantify this cost).
Alternative 6: Push Back Demand
Alternative 6 creates more slack by delaying some customers’ orders into a future month. This option is not ideal and highly discouraged because, though it
allows TSMC to eventually meet its customers’ demand, it does not do so in a timely manner, causing delays in its clients’ processes. Additionally, TSMC is a
market leader because of its reliable and timely service; changes to this may cause their customers to perceive them negatively. Admittedly, there may come
times where delays are inevitable (maybe due to macro factors) and so under those circumstances, TSMC will have to prioritize its clients and make sure to
keep an open and transparent line of communication.
Category 3: Creating policies around emergency orders
Alternative 7: Late-change fees
Late-term design changes are really big inconveniences for TSMC because it puts a big strain on their available resources and disrupts pre-determined
production schedules. To discourage customers from making last-minute requests, and to ensure that the company is properly compensated for their trouble,
we propose that TSMC should charge a late-change fee based on the complexity of amendment.
RECOMMENDATIONS
As we believe that TSMC should avoid pushing back demand as much as possible, we propose that the company uses a mix of the alternatives discussed
above to ensure that they meet their customer’s demand in a timely manner.
We recommend that the delivery of the new CU ECP (only 1) and Photo 183
(only 1) should be expedited to arrive in time to help meet the June/July demand surge; 1,500 demand hours (i.e. 300 N65 hours and 1,200 N90 hours) be
pulled forward from August to July; and in August, that TSMC should rebalance manufacturing capacities by transferring a demand load of 0.9k from N65
to N90
. Though our recommendations would allow TSMC to optimize the use of its resources to ensure that demand is met, it leaves Fab 12 with no slack
capacity, and thus unable to absorb any potential increases in N65 demand or new emergency orders. As such we propose that
TSMC initiates the
qualification process for Fab 14
to ensure that they reserve some “safety capacity”. Finally, as last-minute modifications can cause large disruptions within
the company, we also encourage
TSMC to charge “late-change fees” based on the difficultly of the modification
in an attempt to limit big last-minute
changes and to have some additional resources which they can put towards accommodating this request.
APPENDIX
Appendix 1 - Planning for the Month of August
Appendix 2 - Planning for the Month of June
Your preview ends here
Eager to read complete document? Join bartleby learn and gain access to the full version
- Access to all documents
- Unlimited textbook solutions
- 24/7 expert homework help
Appendix 3 - Planning for the Month of July