T 2263/22 (Assignments of boosters to computation nodes/PARTEC) 30-04-2025

Admissibility of opposition and intervention

1. The boards concurs with the opposition division that the opposition was admissible. The proprietor has not challenged this in the appeal proceedings.

2. The board considers the intervention filed by the intervener to be admissible. This has not been challenged by the proprietor.

Request to postpone the oral proceedings

3. With letter of 27 April 2025, the proprietor requested postponement of the oral proceedings scheduled for 30 April 2025 due to the sudden illness of its "lead counsel". With a second letter on the next day, it was explained that he had been involved in the drafting, examination and opposition phases and was involved in parallel infringement proceedings against the intervener. A medical certificate was provided.

4. The opponent and the intervener indicated on 29 April 2025 in telephone conversations with the registry that they were against a postponement (see notes of these telephone conversations communicated to all parties on the same day).

5. The board decided to not postpone the oral proceedings for the following reasons.

5.1 According to Article 15(2) RPBA, a request of a party for a change of the date fixed for oral proceedings may be allowed if the party has put forward serious reasons which justify the fixing of a new date. If the party is represented, the serious reasons must relate to the representative.

5.2 There were pending proceedings before the UPC and the intervener was against a postponement. The registered representative was not the person referred to as "lead counsel". Even though the participation of the latter was indeed announced with letter of 15 April 2025, his role as "lead counsel" was not apparent from the file. Furthermore, the issues to be addressed at the oral pro­ceedings were only Articles 100(c)/123(2) and 123(3) EPC and did not involve the consideration of any prior art.

6. The board notes that at the oral proceedings the proprietor was represented by two professional representatives. Neither indicated to have had any difficulty in presenting the proprietor's case.

Main request - Article 100(c) EPC

7. The main request concerns the patent as granted.

8. Features 4 to 4.4.3 of granted claim 1 were not inclu­ded in original claim 1 or in any other original claim.

9. The patent was opposed inter alia on the ground that granted claim 1 extended beyond the content of the application as filed, Article 100(c) EPC, in view of features 4.1, 4.2 and 4.4.3. The opposition division agreed with the objection in respect of features 4.2 and 4.3.3 but not in respect of feature 4.1.

10. For the reasons given in the following, the board con­si­ders that granted claim 1 goes beyond the content of the application as filed in respect of feature 4.2. The objections against the other features may be left open.

11. Claim 1 - Feature 4.2 - Interpretation

11.1 Feature 4.2, in combination with feature 4, reads:

"wherein the plurality of computation nodes and the plurality of boosters are arranged such that during processing of said computation task assignments of computation nodes and boosters can be provided such that at least [...]

(ii) one or more of the boosters is shareable by more than one computation node of the plurality of computation nodes"

In the following, whenever feature 4.2 is mentioned, it is understood to be read in combination with feature 4.

11.2 Feature 4.2 defines two alternatives, depending on whether one, or more than one, of the boosters is/are being referred to.

11.3 First alternative: one of the boosters

11.3.1 The board considers that claim 1 requires the plurality of computation nodes and the plurality of boosters to be arranged such that, during processing of a computation task, a booster can be used (shared) by more than one computation node (it being left open whether sequentially or simultaneously) by means of the provision of more than one assignment ("assignments") of computation nodes and boosters during processing of said computation task (which assignments do not include an assignment at the start of the processing of said computation task).

11.3.2 The opposition division had interpreted feature 4.2 as defining, as a property of boosters, that they can be shared between two or more computation nodes during processing of a computation task, "shared" being understood as being simultaneously used by the two or more computation nodes. The proprietor seems to have initially adopted this position but later changed it.

The board agrees with the proprietor's later position that "shareable" in feature 4.2 does not imply the possibility for a booster to be, at some point, used simultaneously by more than one computation node: the booster could also be "shared" and only be used sequentially by more than one computation node during the processing of the computation task, as claim 1 does not require a booster to be shared by means of a single assignment - see "assignments" in feature 4.

11.3.3 The proprietor argued that "shareable" in feature 4.2 only referred to the possibility for any computation node to access any booster via a common communication infrastructure IN, and referred in this respect to paragraph [0037] of the patent.

The board considers that if that were the case, feature 4.2 would be redundant for not going beyond what is already specified by features 1.1, 2 and 4.1. The board rather considers that feature 4.2, read in combination with feature 4, specifically refers to what is achievable - due to the arrangement of the boosters and computation nodes - by "assignments of computation nodes and boosters" provided during processing of the computation task.

Paragraph [0037] of the patent comprises the statement "hence, each of the boosters B can be shared by any of the computation nodes CN", where "hence" refers to the previous statement that "a flexible coupling of boosters to computation nodes is established by a communication infrastructure IN". The board understands this paragraph as explaining that the invention enabled the use of any of the boosters by each of the computation nodes, unlike the acknowledged prior art that involved a fixed coupling of boosters to computation nodes (paragraphs [0007] and [0008]). This a general teaching that does not address the specific situation addressed in feature 4, namely what can be done "during the processing of [a] computation task".

11.3.4 The proprietor also argued that the "assignments" referred to in feature 4 could be static or dynamic assignments. The case of a static and a dynamic assignment was thus encompassed.

The board disagrees. Claim 1 makes a distinction between the provision of an assignment "at a start of processing" and "during said processing" of the computation task: see features 4.4, 4.4.1 and 4.4.3.

This distinction is also consistently made throughout the description. See in particular paragraph [0011]: "The resource manager may establish a static assignment at start of a processing of a computation task. Alternatively or additionally it may establish a dynamic assignment at runtime, which means during processing of the computation task.".

The board therefore interprets "during processing of said computation task" in feature 4 as not including "at start of processing of said computation task". The "assignments" referred to in feature 4 must thus be dynamic assignments, performed at runtime.

11.4 Second alternative: more than one of the boosters

11.4.1 The board considers that the skilled person would interpret the "is" in "one or more of the boosters is shareable by more than one computation node" in feature 4.2 as implying that the "one or more of the boosters" are each "shareable by more than one computation node".

Consequently, the second alternative in feature 4.2 is understood as specifying that more than of one of the boosters are each shareable by more than one computation node.

11.5 The proprietor argued that feature 4.2 defined two alternatives: "one of the plurality of boosters is shareable by more than one computation node ..." and "more than one of the boosters are shareable by more than one computation node ...", the second alternative being understood as a collective shareability, as in paragraph [0011] of the patent.

The use of the word "or" in "one or more" in feature 4.2 emphasised that two alternatives were defined. The wording "at least one" had intentionally not been used. If it had been the intention to say that each of the one or more boosters is shareable, the feature would have been formulated e.g. as "each of the one or more boosters is ...".

The board is not convinced by these arguments.

The board agrees with the proprietor that "or" suggests two alternatives. These alternatives are "one of the boosters is shareable by more than one computation node" and "more [than one] of the boosters is shareable by more than one computation node". In the second alternative, however, the use of the singular verb "is" requires interpretation, since it is applied to a subject that is plural.

The board considers the more natural grammatical interpretation of "is" in that second alternative to be "are each" rather than a collective "are", as also argued by the intervener. It is also a technically meaningful interpretation: it requires the plurality of computation nodes and the plurality of boosters to be arranged such that it is possible, by means of assignments provided during processing of the computation task, to have more than one of the boosters being each used by more than one computation node during processing of said computation task.

The teaching of paragraph [0011] of the patent ("a sharing of accelerators, here in form of boosters, by computation nodes is feasible") is essentially the same as that of paragraph [0037], which has been addressed at point 11.3.3 above. It is not concerned with the specific situation addressed in feature 4, namely what can be done "during the processing of [a] computation task".

The board also notes that neither the formulation "is shareable" nor "is shared" in respect of "one or more" boosters is used in the description.

12. Claim 1 - Feature 4.2 - Added matter

12.1 The board considers that at least the first alternative (one of the boosters) is not directly and unambiguously derivable from the application as filed (represented hereinafter by the PCT publication WO 2012/049247 A1).

12.2 The proprietor has argued, based on several passages of the application as filed, that it was originally disclosed that a booster may be simultaneously or sequentially used by more than one computation node during the performance of a computation task.

The board is not convinced by the proprietor's arguments for the following reasons.

12.2.1 On page 3, lines 8-9, the statement that "a sharing of accelerators, here in the form of boosters, by computation nodes is feasible" would be understood by a skilled person in the sense that, because of the "loose coupling" in the invention instead of the "fixed coupling" in the prior art, a booster may be used not only by a single computation node (to which it would be attached in the prior art) but by any computation node via the communication interface. This neither implies that a booster can be simultaneously used by more than one computation node nor that it can be sequentially used by more than one computation node during the performance of a (hence the same) computation task.

The same understanding applies to the statement that "each of the boosters B can be shared by any of the computation nodes CN" on page 9, lines 10-17, as well as to the statement on page 11, lines 34-35, that "a sharing of accelerator capability between computation nodes becomes possible" as a result of the flexible connection of the computation nodes with the boosters.

12.2.2 On page 3, lines 11-14, the statements that "the resource manager may establish a static assignment at start of processing of a computation task" and that "alternatively or additionally it may establish a dynamic assignment at runtime, which means during processing of the computation task" does not imply that a same booster that has been assigned to a computation node at start of processing may be reassigned, during the processing of the computation task, to another computation node for the performance of (another part of) the same computation task.

Furthermore, as indicated at point 11.3.4 above, feature 4.2 requires the sharing to be the result of more than one assignment during processing of the computation task, which would correspond here to more than one dynamic assignment of the same booster to different computation nodes during the performance of a same computation task, which is not disclosed.

Similar considerations apply to page 4, lines 27-29, and page 7, lines 26-29.

12.2.3 On page 4, lines 2-3, the statement that "it is also possible for the computer cluster arrangement to solve the parts of the computation task in parallel or in succession" does not specifically refer to the solution of these parts by a same booster. It refers the solution of these parts by the plurality of computation nodes and the plurality of boosters in the "computer cluster arrangement".

12.2.4 The appellant noted that figure 2 shows more computation nodes than boosters and argued that if a booster could only be used by a single computation node at a time, the disadvantages of the prior art (not making full use of resources) would not be overcome and the invention would be even less advantageous that the prior art, which provides each computation node with a (fixed) booster. The skilled person would therefore infer that in the invention a booster may be simultaneously used by more than one computation node.

The board is not convinced by this argument. That there may be more computation nodes than boosters in the computer cluster arrangement (as e.g. in figure 2) does not imply that it must be possible to use a booster simultaneously for two (or more) computation nodes. For instance, some computation tasks may be efficiently performed by a computation node without requiring the use of a booster. Such an arrangement is also not necessarily disadvantageous compared to the prior art arrangement (in which boosters are fixedly coupled to respective computation nodes), as it may avoid to have unused boosters. With the invention, the number of boosters relative to the number of computation nodes may be adapted to the application needs, thereby reducing resource waste and avoiding "over- or undersub­scription" of boosters (see page 1, lines 23-26, and page 2, lines 10-12). This does not rest on a possible simultaneous use of a booster for more than one computation node.

12.2.5 On page 7, lines 18-24, it is explained that "processors being applied in computation nodes differ in their processor design compared to processors being applied in boosters" in particular in that the former comprise an "extensive control unit" as "several computation tasks have to be processed simultaneously" by a computation node, whereas the latter only comprise a "simple control structure". The board considers that in view of this explanation the skilled person would assume that, in the context of the invention, no assignments will be provided by which boosters will have to simultaneously process several computation tasks and, similarly, several parts of a computation task provided by two or more computation nodes (as these parts of computation tasks could be as distinct and complex as distinct computation tasks).

As noted by the proprietor, the original description provides an SIMD processor, a GPU or many-core pro­cessors (such that Intel's many-core processor Knight's Corner (KC)) as examples of processors that may be used for the boosters. The skilled person knows that these processors have parallel processing capabilities and would understand that they are expected to be used to efficiently process a part of a computation task provided by a computation node, achieving thereby the desired "boost" or "acceleration". The use of such processors in boosters does however not hint at a simultaneous use in the context of the invention of a booster for processing several parts of a computation task provided by more than one computation node.

12.2.6 The appellant pointed to the reference to a "number of threads" on page 10, line 24, as disclosing a simultaneous use of a booster. The board is not convinced. A booster, like a GPU, may have parallel computation capabilities and the number of threads on page 10 may refer to parallel threads used for the computation of a part of the computation task. There is no disclosure that the different threads would corresponds to parts of the computation task provided by more than one computation node.

12.2.7 An assignment of a single booster to "more than one computation node" is not explicitly disclosed in the application as filed.

As noted by the opponent, several passages of the original description refer to an assignment of at least one booster to one of the plurality of boosters (e.g. page 5, line 23; page 6, lines 20-21 and 23-24; page 12, lines 29-30; page 13, lines 3 and 15; original independent claim 11, line 9), which is coherent with the suggested interpretation.

Other passages refer to an assignment of "at least one booster" to "at least one computation node" (e.g. page 12, line 32, original independent claim 1, lines 8-9, replicated on page 3, lines 1-2). The board considers that the skilled person would understand them - in the light of the overall disclosure of the original application as summarised so far - as referring to a general assignment of boosters to computation nodes, without directly and unambiguously disclosing an assignment in which a single booster is assigned to more than one computation node.

12.2.8 During the oral proceedings, the appellant referred in particular the following passage on page 13, lines 18-21, relating to the embodiment of figure 5:

"Hence, the booster B returns the computed result back to the computation nodes CN. The computation nodes CN may use the returned value for computation of further computation tasks and may again forward at least a further part of a computation task to at least one of the boosters B."

The board considers that it does not directly and unambiguously disclose that the booster that returned the computed result to the computation node (and had thus previously been assigned to it) would be one of the "at least one of the boosters B", nor that any of the "at least one of the boosters B" is one that had previously been assigned to a computation node. Hence, it does not disclose that a same booster is used by more than one computation node during the performance of a same computation task.

13. Consequently, the ground of opposition under Article 100(c) EPC prejudices maintenance of the patent as granted. The main request is not allowable.

Auxiliary request 3A

14. During the oral proceedings, the proprietor asked auxiliary request 3A to be the next request to be discussed (even though auxiliary requests 9.A.0 and 9.A.VI were higher in rank in the proprietor's list of requests).

15. Claim 1 of auxiliary request 3A differs from claim 1 of the main request in that feature 4.2 has been amended as follows:

"(ii) [deleted: one or more of the] boosters [deleted: is] are shareable by more than one computation node of the plurality of computation nodes, and".

16. Admittance

16.1 Auxiliary request 3A was filed in the opposition proceedings (on 26 November 2021) but was not admitted by the opposition division (contested decision, point 47). The proprietor has not argued in the appeal proceedings that the decision not to admit it suffered from an error in the use of the discretion of the opposition division. Its admittance was thus at the discretion of the board under Article 12(6) RPBA, first sentence.

16.2 The opponent and the intervener did not request that auxiliary request 3A not be admitted but instead raised an objection under Article 123(3) EPC against it.

16.3 Compared to auxiliary request 9.A.0, which has been considered in the contested decision and is part of the appeal proceedings, auxiliary request 3A comprises only one of the two amendments made to claim 1 in that request, namely the one aimed at addressing the objection against feature 4.2. Hence, auxiliary request 3A did not raise any new issue beyond those that had to be considered in respect of auxiliary request 9.A.0. The board therefore used its discretion under Article 12(6) RPBA to admit auxiliary request 3A.

17. Claim 1 - Amended feature 4.2 - Article 123(3) EPC

17.1 The proprietor argued that the amendment made to feature 4.2 ("boosters are shareable ...") merely limits claim 1 to the second alternative in feature 4.2 in granted claim 1 ("one or more of the boosters is shareable...") and is therefore compliant with Article 123(3) EPC.

17.2 However, as explained in section 11.4 above, the board interprets that second alternative in granted claim 1 as requiring the more than one boosters to be each "shareable". In contrast, amended claim 1 only requires them to be collectively "shareable", which results in a broader scope.

17.3 The proprietor's arguments against this conclusion focused on the interpretation of the second alternative in feature 4.2 of granted claim 1. They have been addressed in section 11.4 above.

17.4 Claim 1 has therefore been amended in a way that extends the protection conferred by the patent, which is contrary to Article 123(3) EPC.

Further auxiliary requests

18. At the oral proceedings, the proprietor conceded that in view of the position taken by the board on the main request and auxiliary request 3A, none of the other auxiliary requests (including auxiliary requests 9.A.0 and 9.A.VI) would be suitable to overcome both the objection under Articles 100(c)/123(2) EPC and the objection under Article 123(3) EPC.

19. Consequently, the question of the admittance of these other auxiliary requests (which had been raised by the opponent against some of them) may be left open as they are all in any case not allowable.

Request to stay the proceedings until issuance of G 1/24

20. With letter of 15 April 2025 as well as during the oral proceedings, the proprietor requested the appeal proceedings to be stayed until issuance of decision G 1/24. The proprietor argued that the board's interpretation of claim 1 did not take the description and figures into account, as required by Article 69(1) EPC. The outcome of the present case dependent thus on the Enlarged Board's answer to the questions referred in T 439/22, in particular question 2.

The opponent and the intervener were against a stay of proceedings.

21. The board decided not to stay the proceedings for the following reasons.

21.1 The pertinent considerations in T 2116/22 (point 2.4 of the reasons) are followed: The provisions in the EPC concerning a stay of proceedings following a referral to the Enlarged Board of Appeal only concern the referring Board (Article 112(3) EPC). There is, however, no legal basis in the EPC nor in the Rules of Procedure of the Boards of Appeal requiring that any other board stays its proceedings to await the outcome of the proceedings before the Enlarged Board of Appeal. The decision whether or not to stay the proceedings in such cases is thus a discretionary one.

21.2 The questions referred in T 439/22 to the Enlarged Board of Appeal were limited to claim interpretation "when assessing the patentability of an invention under Articles 52 to 57 EPC" (see question 1), whereas in the present appeal case only Articles 100(c)/123(2) and 123(3) EPC were at issue.

21.3 Furthermore, the board did consider the description and figures in its interpretation of claim 1, as is apparent from section 11 above.

21.4 Therefore, the board exercised its discretion not to stay the proceedings.

Apportionment of costs before the first instance

22. According to the minutes of the second oral proceedings (which took place on 28 and 31 January 2022) before the opposition division, dated 8 August 2022:

- the opponent requested during the oral proceedings "apportionment of costs which have been incurred during the oral proceedings" (point 10.1),

- the opposition division concluded "that the request of apportionment of costs is granted, and that the costs will be fixed later" (point 10.6)

- the chairperson announced as decision of the opposition division that "the request of apportionment of costs filed by the opponent is granted" (point 11.2).

23. In the written decision, the opposition division indicated the following:

- "the opponent requested apportionment of costs and argumented [sic] it as following: [...] The big number of the requests imposed a big quantity of preparation work. [...]" (point 51),

- "The unnecessarily high number of requests lead to an unnecessary longer preparation time for the opponent. Also filing some requests later results in repeating the process of consulting with the client and taking the client's instructions on them just for the additional requests which would inevitably have lead to additional and avoidable costs.

Taking into account also that for the oral proceedings of 28.01. and 31.01.2022 there were 40 (3A 3F, 5.I - 5.IV, 7.A.0 - 7.F.IV) out of 70 requests that lead to unnecessary expenses, the opposition division decides that the proprietor should bear half of the costs incurred by the opponent when preparing the oral proceedings of 28.01.2022 and 31.01.2022.

It is noted that this refers to different costs incurred for preparing the oral proceedings and not to the costs incurred by the mere presence at the oral proceedings" (point 53, last three paragraphs),

- "Regarding the request of apportionment of costs, the division decides that the proprietor should bear half of the costs incurred by the opponent when preparing the oral proceedings of 28.01.2022 and 31.01.2022." (point 54.2),

- "It is noted that this refers to the different costs incurred for preparing the oral proceedings and not to the costs incurred by the mere presence at the oral proceedings" (point 54.2.1).

24. The proprietor argued that the decision taken by the opposition division at the second oral proceedings ("first decision") was, according to the minutes, to accept the request formulated by the opponent during the oral proceedings, i.e. an apportionment of those costs "which have been incurred during the oral proceedings" (emphasis added).

In the written decision, the opposition division formulated instead a different, "second decision": an apportionment of "half of the costs incurred by the opponent when preparing the oral proceedings of 28.01.2022 and 31.01.2022" (emphasis added).

According to the proprietor, both decisions were to be set aside. The first decision had not been reasoned in the written decision. The second decision concerned a request which had never been submitted nor had it been discussed in the oral proceedings (statement of grounds of appeal, page 35). The proprietor's right to be heard had been infringed in that respect (letter of 20 July 2023, paragraph bridging pages 3-4).

The proprietor then gave reasons as to why none of these two decisions were justified. There was in particular no justification for an apportionment of costs incurred during the oral proceedings. It had already been agreed at the end of the first oral proceedings that the second oral proceedings would be scheduled for two days, as was then done with the summons dated 15 October 2021, hence before the proprietor's submissions on 20 January 2022 (statement of grounds of appeal, pages 35-37; letter dated 20 July 2023, pages 3-5).

25. The opponent argued that its request during the second oral proceedings comprised in effect two requests: a first request for an apportionment of costs, and a second request for a calculation of the amount of those costs which had incurred during the oral proceedings. During the oral proceedings, the opposition division agreed to the first request and did not give any indications regarding the second request. The opponent then argued why the number of submitted requests was excessive and constituted an abuse of procedure by the proprietor (letter of 29 April 2013, pages 4-6).

26. The board agrees with the proprietor as regards the procedural defects of the first-instance proceedings.

26.1 It is not apparent from the minutes of the second oral proceedings that the opponent's request was formulated as two distinct requests. From these minutes, it is to be concluded that the opposition division has decided to grant an apportionment of costs incurred during the oral proceedings in favour of the opponent.

This decision has not been reasoned in the written decision. On the contrary, the written decision explicitly indicates that such costs should not be apportioned in favour of the opponent (contested decision, points 53, last paragraph, and 52.1).

Hence, the decision as notified to the parties in respect of the apportionment of costs deviates substantially from the decision taken at the oral proceedings, thereby violating Rule 111(1) and (2) EPC. In accordance with established case law (see e.g. T 425/97 and T 1698/06), this discrepancy amounts to a first substantial procedural violation.

26.2 The written decision concerning "half of the costs incurred by the opponent when preparing the oral proceedings of 28.01.2022 and 31.01.2022" (contested decision, point 52) and the reasons given for it (contested decision, point 53) have been conveyed for the first time to the proprietor with the written decision. The proprietor had thus no opportunity to comment on them, contrary to its right to be heard, Article 113(1) EPC. This is a further substantial procedural violation.

26.3 The written decision on the apportionment of costs is thus to be set aside.

27. The opposition division's decision on the apportionment of costs is also wrong in substance.

27.1 The arguments of the opponent and the considerations of the opposition division only turn on the auxiliary requests that were filed by the proprietor but fail to explain why the proprietor's conduct during the oral proceedings would justify that the costs incurred during the oral proceedings be borne by the proprietor, as requested by the opponent.

According to the minutes of the second oral proceedings, only 5 claim requests were effectively discussed orally: auxiliary requests 9.A.0, 9.A.I, 9.A.V, 9.A.VI and 9.A.VIII, the last two being filed by the proprietor during the oral proceedings (minutes, point 5.3). The many other auxiliary requests were summarily not admitted in the proceedings after the discussion of these five requests (minutes, point 9).

In view of these circumstances, the board fails to see an abuse of procedure in the behaviour of the proprietor during the oral proceedings.

27.2 The board also considers that neither the high number of auxiliary requests nor the circumstances of their filing makes equitable an apportionment of costs for the preparation of the oral proceedings in favour of the opponent, Article 104(1) EPC.