Predictika · Semiconductor manufacturing equipment

An AI sales agent's fab tool order, checked before it goes out

An AI sales agent prepares an order for a Tokyo Electron (TEL) coater/developer, etch system, single-wafer cleaning system or batch thermal processing system for a 300 mm logic fab: the tool family and model, wafer size, process, capacity, substrate and whether the tool is new. Before it goes out, the agent sends it to Predictika, which checks it against a declarative model of the comparison tables published on tel.com and returns a verdict, every reason and the table figure it computed. The customer is fictional.

Scope: TEL coater/developers, etch systems, single-wafer cleaning systems and batch thermal processing systems: family, model, wafer size, process, throughput, chambers, batch load, temperature, substrate and availability. See everything this validator covers

This validator models orders of 17 TEL tools, based on the comparison tables on tel.com, not just the scenarios shown below.

  • Coater/developers — LITHIUS Pro DICE, Pro AP, Pro Z, Pro V, Pro and LITHIUS: the wafer sizes, the processes each lists (EUV, immersion, ArF, KrF, i-line, spin-on carbon, polyimide, package, spin-on dielectric and OCCF) and the throughput each table states; the LITHIUS is listed as certified used only.
  • Etch systems — Tactras, Episode UL and Certas on 300 mm wafers: the processes each lists (dielectric, conductor, RIE, chemical dry etch) and the chamber limit each table states.
  • Single-wafer cleaning — CELLESTA i MD, CELLESTA i and CELLESTA MS2: pre- and post-clean, wet etch and physical clean, the throughput each table states and the chamber limit where one is stated; where it is not, a person confirms the chamber count.
  • Batch thermal processing — EVAROS, TELINDY PLUS, TELFORMULA, TELINDY PLUS IRad and ALPHA-8SE i: wafer sizes, processes, the production load per batch and the temperature range each table states.
  • What the order says about itself — the family against the model, the substrate (silicon only outside the coaters, where a person confirms any other) and the capacity figure the agent writes into the order, compared with the table.

Any order of these TEL tools that touches these rules is in scope — not only the scenarios pictured here. Not covered: other TEL products, prices (TEL does not publish them), delivery, site and utility requirements, service contracts, and process results beyond the tables.

The AI sales agentBuilds the tool order from the conversation and says it is ready.
Predictika validatorChecks every rule at once and returns the verdict, each finding, and the computed figures.

8 scenarios in 3 groups · each group runs from the agent’s first attempt to the outcome · click one to switch

What the agent proposes

The AI sales agent says

Illustrative agent statement. The validator receives the structured request below.

Proposing agent: AI sales agent (scripted)

Before sending, the agent must call Recorded run

Predictika validator


      

        
        
        
      

      
Waiting for the call.

      
    

  

  
  

    
Where rules-based CPQ gets it wrong
Click to expand
    
Our assessment of what a typical rules-based CPQ (product rules, price rules and validation rules, as in Salesforce CPQ) does with the same cases. It is a judgment about effort and reliability, not a claim that nothing can be built.

    
What the validator caughtTypical rules-based CPQPredictika
A process that depends on the modelOnly with workarounds
Each model lists its own processes, and an order names the process separately from the model. A rules-based CPQ needs a rule per model and process pair, and the process is often not a field at all.		The model and the process arrive in one request and are compared against the table on every call.
Limits that change with the modelOnly with care
Throughput, chamber limit, batch load and temperature range each differ by model and by family. A rule can encode each case, but they must be kept in step with the tables.		The model is read together with the order and the limit that applies is computed on every call.
A capacity the agent statesOnly with care
The stated capacity is a free-entry field; a rule has to know which model and which kind of limit it refers to, then compare.		The stated capacity is a field in the request, compared with the table figure for the model on every call.
Every problem returned in one answerOnly with workarounds
Rules tend to fire as the quote is saved or a step is completed, so a person often fixes one error and meets the next.		All findings come back together, whatever order the choices arrived in.

  

  

    
What this validator catches: findings from its test cases
Click to expand
    
Every rule is checked on every call, in any order the choices arrive; one problem never hides another. Blocking findings stop the quote; approval findings need a person to sign off; everything else is computed.

    
1The LITHIUS Pro DICE lists EUV, immersion, ArF, KrF, i-line and spin-on carbon processes; the process the order needs is not among them.
2The LITHIUS Pro AP lists polyimide, package, spin-on dielectric and OCCF processes; the process the order needs is not among them.
3The LITHIUS Pro Z lists EUV, immersion, ArF, KrF, i-line and spin-on carbon processes; the process the order needs is not among them.
4The LITHIUS Pro V lists immersion, ArF, KrF, i-line and spin-on carbon processes; the process the order needs is not among them.
5The LITHIUS Pro lists immersion, ArF, KrF, i-line and spin-on carbon processes; the process the order needs is not among them.
6The LITHIUS lists immersion, ArF, KrF and i-line processes; the process the order needs is not among them.
7The Tactras lists dielectric, conductor and reactive ion etch processes; the process the order needs is not among them.
8The Episode UL lists dielectric, conductor and reactive ion etch processes; the process the order needs is not among them.
9The Certas LEAGA lists dielectric and chemical dry etch processes; the process the order needs is not among them.
10The CELLESTA-i MD lists pre-clean, post-clean and wet etch processes; the process the order needs is not among them.
11The CELLESTA-i lists pre-clean, post-clean and wet etch processes; the process the order needs is not among them.
12The CELLESTA MS2 lists physical force cleaning processes; the process the order needs is not among them.
13The EVAROS lists oxide, nitride and poly processes; the process the order needs is not among them.
14The TELINDY PLUS lists thermal ALD, LPCVD, oxidation and anneal processes; the process the order needs is not among them.
15The TELFORMULA lists oxide, anneal, nitride, poly, TEOS and high-k processes; the process the order needs is not among them.
16The TELINDY PLUS IRad lists plasma-assisted ALD processes; the process the order needs is not among them.
17The ALPHA-8SE i lists oxide, anneal, nitride, poly, TEOS, HTO and high-k processes; the process the order needs is not among them.
18The model ordered belongs to a different tool family than the one the order names.
19The order has wafers of 200 mm; this model takes 300 mm wafers only.
20The order has wafers of 150 mm; this model takes 200 mm or 300 mm wafers.
21The order has wafers of 300 mm; this model takes 150 mm or 200 mm wafers.
22The order needs 301 wafers an hour; the TEL table gives 300 wafers an hour for this model.
23The order needs 441 wafers an hour; the TEL table gives 440 wafers an hour for this model.
24The order needs 651 wafers an hour; the TEL table gives 650 wafers an hour for this model.
25The order needs 801 wafers an hour; the TEL table gives 800 wafers an hour for this model.
26The order needs 201 wafers an hour; the TEL table gives 200 wafers an hour for this model.
27The order has 7 chambers; the TEL table allows up to 6 for this model.
28The order has 13 chambers; the TEL table allows up to 12 for this model.
29The order has 21 chambers; the TEL table allows up to 20 for this model.
30The TEL table states no chamber count for this CELLESTA model; a person should confirm it.
31The order loads 201 wafers a batch; the TEL table gives a production load of up to 200 for this model.
32The order loads 126 wafers a batch; the TEL table gives a production load of up to 125 for this model.
33The order loads 51 wafers a batch; the TEL table gives a production load of up to 50 for this model.
34The order loads 151 wafers a batch; the TEL table gives a production load of up to 150 for this model.
35The order needs 801 degrees C; the TEL table gives a range up to 800 degrees C for this model.
36The order needs 1101 degrees C; the TEL table gives a range up to 1100 degrees C for this model.

  

  

    
Scenarios and results
Click to expand
    
The recorded result of the validator for each scenario on this page: the request the AI agent sent, the verdict, and every finding, in the order they appear above.

    
1. Batch furnace order: HfO2 anneal on an EVAROS

    
The agent's request. Please quote a TEL EVAROS for a batch thermal processing system for high-k processing on 200 mm non-silicon wafers. Each batch loads 300 wafers at 1000 degrees C. The tool is new. It is ready to send.

    
Verdict: Blocked.

    
    
      
  • Model takes 300 mm wafers only. The order has wafers of 200 mm; this model takes 300 mm wafers only.
    • 
      
  • EVAROS process not listed. The EVAROS lists oxide, nitride and poly processes; the process the order needs is not among them.
    • 
      
  • Batch larger than the furnace takes. The order loads 300 wafers a batch; the TEL table gives a production load of up to 200 for this model.
    • 
      
  • Process temperature above the model range. The order needs 1000 degrees C; the TEL table gives a range up to 800 degrees C for this model.
    • 
      
  • Substrate other than silicon. The TEL table lists silicon substrates only for this tool family.
    • 
    

    
2. Batch furnace order: The agent resubmits after fixing the errors

    
The agent's request. Please quote a TEL TELFORMULA for a batch thermal processing system for high-k processing on 200 mm silicon wafers. Each batch loads 300 wafers at 1000 degrees C. The tool is new. It is ready to send.

    
Verdict: Blocked.

    
    
      
  • Model takes 300 mm wafers only. The order has wafers of 200 mm; this model takes 300 mm wafers only.
    • 
      
  • Batch larger than the furnace takes. The order loads 300 wafers a batch; the TEL table gives a production load of up to 50 for this model.
    • 
    

    
3. Batch furnace order: The agent resubmits after fixing the errors

    
The agent's request. Please quote a TEL ALPHA-8SE i for a batch thermal processing system for high-k processing on 200 mm silicon wafers. Each batch loads 150 wafers at 1000 degrees C. The tool is new. It is ready to send.

    
Verdict: Allowed.

    
    
      
  • Order summary. The TEL table gives this model a production load of up to 150 wafers and a range up to 1250 degrees C.
    • 
    

    
4. Etch chambers: Eight chambers on a Tactras

    
The agent's request. Please quote a TEL Tactras for an etch system for dielectric etch processing on 300 mm silicon wafers. The line has 8 chambers. The tool is new. The quote states a chamber limit of 8 chambers. It is ready to send.

    
Verdict: Blocked.

    
    
      
  • More chambers than the model takes. The order has 8 chambers; the TEL table allows up to 6 for this model.
    • 
      
  • Stated chamber count does not match. The order states a limit of 8 chambers; the TEL table gives up to 6 for this model.
    • 
    

    
5. Etch chambers: The agent resubmits after fixing the errors

    
The agent's request. Please quote a TEL Tactras for an etch system for dielectric etch processing on 300 mm silicon wafers. The line has 6 chambers. The tool is new. The quote states a chamber limit of 8 chambers. It is ready to send.

    
Verdict: Blocked.

    
    
      
  • Stated chamber count does not match. The order states a limit of 8 chambers; the TEL table gives up to 6 for this model.
    • 
    

    
6. Etch chambers: The agent resubmits after fixing the errors

    
The agent's request. Please quote a TEL Tactras for an etch system for dielectric etch processing on 300 mm silicon wafers. The line has 6 chambers. The tool is new. The quote states a chamber limit of 6 chambers. It is ready to send.

    
Verdict: Allowed.

    
    
      
  • Order summary. The TEL table allows up to 6 chambers on this model.
    • 
    

    
7. Coater process: EUV on a packaging coater

    
The agent's request. Please quote a TEL LITHIUS Pro AP for a coater/developer for EUV processing on 300 mm silicon wafers. The fab needs 100 wafers an hour. The tool is new. The quote states a throughput of 200 wafers an hour. It is ready to send.

    
Verdict: Blocked.

    
    
      
  • LITHIUS Pro AP process not listed. The LITHIUS Pro AP lists polyimide, package, spin-on dielectric and OCCF processes; the process the order needs is not among them.
    • 
    

    
8. Coater process: The agent resubmits after fixing the errors

    
The agent's request. Please quote a TEL LITHIUS Pro DICE for a coater/developer for EUV processing on 300 mm silicon wafers. The fab needs 100 wafers an hour. The tool is new. The quote states a throughput of 440 wafers an hour. It is ready to send.

    
Verdict: Allowed.

    
    
      
  • Order summary. The TEL table gives this model a throughput of 440 wafers an hour.
    • 
    

  

  

    
Narration transcript
Click to expand
    
The text of the narrated walkthrough, step by step.

    
    
      
  1. Welcome to a demo of Predictika's Validator. An AI sales agent prepares an order for a Tokyo Electron (TEL) coater/developer, etch system, single-wafer cleaning system or batch thermal processing system for a 300 mm logic fab: the tool family and model, wafer size, process, and more. Before it goes out, the agent sends it to Predictika, which checks it against a declarative model of the comparison tables published on tel.com and returns a verdict, every reason and the table figure it computed. The customer is fictional.
    2. 
      
  2. I will start with the first request by the agent.
    3. 
      
  3. The agent says. Please quote a TEL EVAROS for a batch thermal processing system for high-k processing on 200 mm non-silicon wafers. Each batch loads 300 wafers at 1000 degrees C. The rest of the request is on screen.
    4. 
      
  4. Before it goes out, the agent calls the validator.
    5. 
      
  5. Blocked, with five findings, all in one answer. The order has wafers of 200 mm; this model takes 300 mm wafers only. The EVAROS lists oxide, nitride and poly processes; the process the order needs is not among them. The order loads 300 wafers a batch; the TEL table gives a production load of up to 200 for this model. The other two are on screen.
    6. 
      
  6. The validator returned five blocking findings in one answer. A check that reads one field at a time would pass most of this; it shows only when the choices in the quote are checked against each other.
    7. 
      
  7. What this shows: for a process that depends on the model, a rules-based CPQ can do this only with workarounds.
    8. 
      
  8. Now the agent resubmits.
    9. 
      
  9. The corrected request. It changes the model to TELFORMULA and the substrate to silicon.
    10. 
      
  10. It calls the validator again.
    11. 
      
  11. Still blocked, with two findings, all in one answer. The order has wafers of 200 mm; this model takes 300 mm wafers only. The order loads 300 wafers a batch; the TEL table gives a production load of up to 50 for this model.
    12. 
      
  12. Three findings cleared; still blocked on model takes 300 mm wafers only and batch larger than the furnace takes.
    13. 
      
  13. The agent resubmits once more.
    14. 
      
  14. The revised request. It changes the model to ALPHA-8SE i and the wafers per batch to 150.
    15. 
      
  15. It calls the validator once more.
    16. 
      
  16. This time the validator clears it.
    17. 
      
  17. Both earlier findings are cleared.
    18. 
      
  18. Five problems caught in one call, fixed in two passes, and a clean quote. That is the value of Predictika's validator: catching mistakes before they reach the customer.
    19.