`
`Jéde991,319?“5
`
`Microelectronics Journal, 26 (1995) 431—440
`
`Sea-of-gates
`alchitecture
`
`Manoel E. de Lima1 and David J.
`Kinniment2
`
`IDepartmento de Infirmética, CCEN, Universidade Federal de Pemambuco, Av. Professor
`Luiz Freire, s/n, Cidade Universita’ria, Recife, Pemambuco, Brazil, CEP50.732—970.
`Tel: (+55) (081) 271 8430. Fax: (+55) (081) 271 4925. E-mail: mel@di.ufiae.br
`2Dep'artment QfElectrical and Electronic Engineering, University ofchastle-upon-Tyne,
`Newcastle-upon-Tyne NE 1 7RU, UK. Tel: (+44) (0)191 222 6000. Fax: (+44)
`(0)191 2.22 8180. E-mail: david.kinniment@newcastle.ac.uk
`
`This paper deals with the development of sea-of—gates
`technology for the design of VLSI circuits. Sea-of—gates
`technology, also considered a second generation gate-array
`system, is discussed in detail. Its internal structure and the
`main physical aspects that differentiate sea-of-gates from
`conventional gate-arrays are presented. The advantages over
`the previous generation ofgate-arrays, and for some circuits
`in comparison with full-custom technology, are discussed,
`together with its influence in the likely ASIC market.
`
`1. Introduction
`
`ince computers first began to be used to
`design chips more than 20 years ago, they
`have become essential for the design of complex
`circuits. With the rapid development of highly
`integrated electronic systems, more sophisticated
`software systems have also had to be produced in
`order to keep pace with the evolution of the
`technology. Today, large CAD systems contain
`millions of lines of code for the specification of
`state—of—the—art
`integrated
`circuits. Despite
`increasing complexity, the turnaround time for
`the layout of designs has been reduced from
`months and weeks to days and hours in recent
`CAD systems. This reduction in time is in part
`the result of the development in design styles
`that has
`taken place in the last
`few years,
`prompted by the need for more integration,
`higher density and greater flexibility. Among the
`
`0026—2692/95/$9.50 © 1995 Elsevier Science Ltd
`
`design styles that have been applied to today’s
`Application Specific Integrated Circuit (ASIC)
`projects,
`standard
`cell
`and
`gate—array
`technologies
`are
`good
`examples
`of
`the
`application of CAD in the ASIC industry. Semi-
`custom technology
`has
`recently
`taken
`a
`considerable portion of this market, mainly as a
`result of its new member,
`the sea-of—gates,
`which has had a considerable impact on the
`architecture of the new generation of ASIC
`technology.
`
`2. Semicustom technology
`
`Semicustom technology [1] basically consists of
`wafer pre-processing up to the level of metalliza—
`tion patterning. This means that all transistors in
`the array matrix are pre-defined with fixed size
`and positions. Therefore,
`in general,
`the effi—
`ciency and performance may not be as good as
`that of full—custom designs, in which all devices
`are created specifically for a particular design.
`Full-custom technology [2] can result in higher
`area efficiency, and consequently a better speed
`and power consumption. Nevertheless, because
`of lower development costs and fast prototype
`turnaround time, semicustom methodology is an
`attractive option for many electronic systems
`requiring small or medium volume production.
`
`431
`
`AMD EX1050
`
`AMD v. Aquila
`IPR2019-01525
`
`AMD EX1050
`AMD v. Aquila
`IPR2019-01525
`
`
`
`M.E. de Lima and D.J. Kinniment/Sea-of-gates architecture
`
`Within the semicustom family, a new class of
`gate-array, the sea—of—gates, has received consid-
`erable attention in the last few years because it
`offers more flexibility and higher density than
`conventional arrays. As a result, it has become
`one of the most popular components in this
`class. The
`sea-of-gates
`architecture, which
`employs a complete carpeting of gates in the core
`area of the chip, has now become dominant in
`the implementation of large—scale systems on
`gate—array architectures. Intense competition and
`short product lifecycles have created enormous
`pressure on the time needed for development
`and manufacture, and sea-of-gates has become a
`useful technology for companies which require
`speed, good performance,
`low cost and high
`density in their ASIC designs [3—5].
`
`3. Sea—of—gates
`
`Sea-of—gates (SOG) technology was introduced
`in around 1982 to provide a topology that
`allowed more flexibility in layout between
`channels and gates [6—17]. Considered to be the
`second generation of gates—arrays, sea-of—gates
`arrays can also be called ‘continuous gate arrays’,
`‘channelless arrays’ or ‘gate forest arrays’. The
`technique provides an implementation environ—
`ment with the characteristics of both full custom
`
`and semicustom designs, offering the low cost
`and speed of semicustom gate arrays in fabrica—
`tion and design time [2, 18, 19], and retaining
`some of the density and performance of full
`custom technology.
`
`Since the basic pitch of transistors in silicon is
`different and denser in sea—of—gates
`than in
`conventional arrays,
`this technology requires
`more sophisticated methods to cope with its
`layout. Also, with the possibility of two or
`more metal layers for routing, and the use of
`sub—micron CMOS technology, even more
`dense and integrated systems will be possible,
`demanding more power
`from each of the
`layout tools and better integration between the
`traditional stages of the layout synthesis,
`i.e.,
`
`432
`
`partitioning,
`routing.
`
`floorplanning,
`
`placement
`
`and
`
`This new generation of gate—arrays has promo-
`ted a substantial advance in semicustom design
`in the VLSI system integration world, allowing
`circuits to be built of the order of more than
`
`1,000,000 gates per chip, a number which it is
`estimated may grow to around 20,000,000 by
`the year 2000. Sea—of—gates gives
`a
`similar
`density but greater flexibility compared to that
`achieved in full custom design, mainly because
`it is possible to use almost 100% of the silicon
`area in regular structures such as memories or
`PLAs.
`
`Because of its density and flexibility, the inher-
`ent hierarchy of circuits can be more easily
`exploited. This is a very important feature, since
`a semicustom environment usually has a reduced
`layout
`flexibility when compared to the full
`custom environment. A designer can now
`exploit this flexibility at higher levels, to improve
`the speed of the design and produce more dense
`and regular circuits [7].
`
`Two basic and important concepts are the heart
`of the modern sea-of—gates technology: removal
`of the conventional channels for routing [5, 6,
`19—26] and gate isolation (GIC)[21—23, 27, 28].
`These two characteristics are the keys for the
`density and flexibility of this new member of the
`gate—array family.
`
`3.1 Channelless architecture
`
`Conventional gate-array architecture uses chan—
`nels, which appear as wide gaps between the
`rows, to allow enough space for inter— and intra-
`cell routing. The interconnections are done by
`processing the metal layer patterning of the logic
`gates that comprise the matrix of gates. The
`channels in the gate-array technology can be
`positioned either between the rows or columns,
`as shown in Fig. 1a. In general, two levels of
`interconnections are used for orthogonal track—
`ing (vertical and horizontal routing). A third
`
`
`
`Microelectronics Journal, Vol. 26, No. 5
`
`level can also be used in some cases for power
`and ground. Because of the extra space reserved
`for these routing channels, some silicon is not
`used for active devices. Consequently, there may
`be a low density of circuits on the silicon.
`
`The new concept in gate arrays, called chan—
`nelless, as depicted in Fig. 1b, shows that the
`new generation of gates-arrays does not contain
`pre-defined wiring channels between cells for
`the custom metallizaljon. The whole area of the
`
`die is now filled with potentially active p— and
`n—channel transistors, and the interconnections
`are made by metal layers over unused transis—
`tors. The routing process and device connec—
`tions generally employ two-sided metallization
`in rows and columns that run above the tran—
`
`sistors, and are isolated from them by an insu-
`lating material. These levels are implemented
`with the connections spaced on a grid where
`the centres of the contacts holes (vias) are loca—
`ted on the grid points. The metal layers usually
`have a preferential direction on each level.
`Depending on the technology used, sea-of—gates
`can also provide more than two metal layers for
`routing. For instance,
`in a three metal
`layers
`technology, the first metal layer could be used
`for building up the basic logic elements (intra—
`cell connections), and the second and third
`
`layers to perform the vertical and horizontal
`routing among the logic functions (inter—cell
`connection)
`or
`power
`distribution. This
`arrangement has been used to improve the effi-
`ciency for the router [7, 19]. The new tech-
`nology allows highly integrated circuits to be
`built, offers fast turnaround and good perfor—
`mance.
`
`3.2 Gate isolation
`
`The second important concept, gate isolation,
`consists of replacing the field oxide usually
`providing the isolation between basic cells in a
`conventional gate-array by transistors. Figure 2
`shows a typical NAND gate that is implemented
`on a conventional gate array (Fig. 2a) and on a
`channelless configuration (Fig. 2b).
`
`In a conventional architecture, common gates
`are used to control the data flow on both n and p
`transistors which are grouped in a pre—deter—
`mined number to create the basic cell in the
`
`array. Furthermore, the oxide isolation, which is
`also used in some sea—of-gates architectures, is
`employed to separate the basic cells on the array,
`thus increasing the gate area and giving a low
`density on the chip. However, with the gate
`isolation technique, no gap in diffusion layers
`isolating a cell from the other neighbouring ones
`
`
`
`
`
`DDDDUUDDDDDUNMHEJDDDDDDDDD
`.
`
`D
`B
`DDD
`one
`
`
`
`
`
`
`
`DUDDDU DD
`
`D
`
`
`lI
`
`
`DDDDDDDUDU1LI
`
`
`
`---DDDDDDDDDUDDD
`
`
`UDDDDUUUD El
`
`
`
`
`
`
`
`rowcelapproach
`
`
`
`
`DU
`DDDDDUUDDD
`
`
`
`
`Fig. l. Gate-array architecture. (3) Conventional gate-array architecture; (b) sea-of—gates architecture.
`
`433
`
`
`
`UDDDUUUDDDDU
`
`
`
`ME. de Lima and D.J. Kinniment/Sea-of-gates architecture
`
`(3) SehemaflcofhlayomothANDgfle
`
`massed-cabs
`
`(b)
`
`l1!" WEE
`
`ll-II Il-ll Em
`IJ
`
`ScharMIIcoimehyoulofaNANDgneb/
`acorwamlondhaslccdcorwwm
`MW“
`
`(a) Conventional gate—array architecture; (b) sea—of—gates archi-
`Fig. 2. Conventional and sea—of—gates arrays architecture.
`tecture.
`
`exists, and continuous p and n transistor chains
`can be created in the silicon die without inter—
`
`and achieving a significantly reduced final area
`[8, 29].
`
`ruptions (Fig. 2b).
`
`Gate isolation is only provided where necessary,
`and it is done by a transistor gate connected to
`the appropriated power
`line (V55 or VDD),
`setting the respective transistors in a cut—off state.
`Hence, the transistors are isolated to help create
`the circuit required. The main advantages of this
`technology are:
`
`0 significant
`density,
`
`increase
`
`in transistor packing
`
`0 less wiring when creating complex functions,
`and
`
`0 easy control of the data flow.
`
`Moreover, because the transistors have diEerent
`gates, these gates can be used separately in order
`to implement different logic, or for facilitating
`the routing to critical areas of the chip.
`
`these two
`In recent sea—of—gates architectures,
`techniques, channelless layout and gate isolation,
`have been used together,
`thus improving the
`flexibility, density and routability of the circuits
`
`434
`
`4. Sea-of—gates styles
`
`In a sea-of-gates architecture, the macrocells are
`composed by repeating basic cell boundaries
`horizontally or vertically on either
`a Row
`Macrocell
`(RMC) approach (Fig. 3a) or
`a
`Column Macrocell (CMC) approach (Fig. 3b)
`[29, 30] until enough gates are assembled. These
`two techniques provide different feed—through
`capabilities and difierent penalties for the vertical
`and horizontal routing. The terms ‘macrocell’
`and ‘functional cell’ are here used to designate
`large cells with a certain logic function whose
`shape and area must be considered in the place-
`ment of the cells in the array.
`
`4.1 The RMC architecture
`
`is
`the basic cell
`With the RMC architecture,
`usually represented by a single vertical pair or
`pairs of transistors, as depicted in Fig. 3a. The
`macrocell in this architecture is implemented by
`repeating the basic cell along the horizontal
`direction until enough gates are allocated to the
`macrocell. In this technique, primitive cells such
`as gates NAND, XOR, etc., which are placed
`closely to create macrocells, do not have,
`in
`
`
`
`Microelectronics Journal, Vol. 26, No. 5
`
`
`
`
`
`Fig. 3. Sea-of—gates architectures. (a) Layout of a macrocell using conventional gate isolation sea—of—gates in a Row Macrocell
`(RMC) approach; (b) layout of a macrocell using sea—of—gates in a Column Macrocell (CMC) approach; (c) layout of a
`macrocell using gate isolation with uniform power distribution.
`
`general, enough feed—through to provide the
`intra—cell connections and so guarantee the
`routability of the circuits. As a result of this
`strategy, horizontal wiring channels are required
`to supply extra free tracks for routing. The allo—
`cation of firee tracks in the congested area varies
`by one basic cell (BC) row pitch in this struc—
`ture. This number of free tracks per row, which
`depends upon the technology used, can repre-
`sent an increment of 13, 15, 20 tracks, etc. In the
`worst case, one BC row is allocated for each
`single track shortage, causing a low transistor
`utilization. This architecture does not have the
`
`capacity to distribute mesh power because of its
`variable—width structure horizontally and an
`insufficiency of feed-through tracks. Power
`distribution is
`therefore needed among the
`macrocells, which promotes difficulties in the
`routing, as illustrated in Fig. 3a.
`
`To increase the gate utilization, a low increment
`in the number of tracks per basic cell is needed.
`Finally, using the RMC architecture,
`the
`arrangement of big cells such as memories and
`large macrocells is restricted to being located
`between the power buses.
`
`4.2 The CMC architecture
`
`On the other hand, in the CMC architecture the
`basic cells can have different sizes and also
`
`different sets of p and n transistors involving
`different numbers of transistors, as shown in Fig.
`3b. The number of transistors must be well
`
`chosen in order to allow the ready implementa-
`tion of gates in the system library.
`
`In this approach, the macrocells are in general
`created by joining basic cells vertically, until
`enough gates are gathered together to form
`the macrocells. Because of the regularity of
`the basic cell distribution in a CMC archi—
`
`is possible to
`it
`tecture on the silicon die,
`have a fine mesh power distribution in the
`array [31].
`In this architecture,
`the wiring
`channels, created in a congested area,
`run
`vertically
`along
`the
`gates,
`providing
`the
`numbers of tracks needed for
`routing. The
`Width of the channels,
`in contrast
`to the
`RMC, can change smoothly by the width of
`one column of pn transistors to provide the
`routing tracks, hence eliminating a big waste
`of silicon, and consequently providing a high
`utilization of gates in the circuit.
`
`435
`
`
`
`ME. de Lima and DJ. Kinniment/Sea-of-gates architecture
`
`4.3 An alternative architecture
`
`Although both of the above architectures use
`gate isolation, the idea of a completely channel-
`less layout seems not to be 100% implemented.
`Another architecture oflers rather more flex—
`
`ibility in the design, as depicted in Fig. 3c. In this
`structure,
`the macrocells Can grow in both
`directions without restriction in the routing of
`the power supply, since it is distributed along the
`circuit using a second level of metal when
`needed. In this architecture, the isolation gate
`approach can be totally exploited, since there is
`always
`a continuous
`sequence of transistors
`without limits for all macrocells. Despite the
`advantages and disadvantages in different sea—of-
`gates architectures, a good result
`in the final
`layout also depends critically upon the algo—
`rithms used for partitioning, placement and
`routing [7, 25, 28, 32—34] of the circuits and the
`number of metal layers availbale for routing.
`
`5. Microarchitectures and
`macroarchitectu res
`
`In terms of the internal structure of basic cells
`
`and their distribution on the silicon, sea—of—gates
`
`two levels;
`can be classified hierarchically at
`namely
`microarchitectures
`and
`macro-
`architectures.
`
`5.1 Microarchitecture
`This level describes the internal characteristics of
`
`the core cells, which are used to build the overall
`structure of the array. These core cells (often
`basic cells) which are a combination of a certain
`number of p and n transistors, can be connected
`in different configurations, depending on the
`technology adopted.
`
`A good example of a simple sca—of—gate’s struc-
`ture using gate isolation is that developed at Delft
`University, called Fishbone [8]. This structure is
`very similar to most of the sea—of—gates basic cells.
`This particular architecture uses gate—isolation in
`a 1.6 pm CMOS process with two layers of metal
`for routing. As depicted in Fig. 4a, each basic cell
`is composed of a pair of transistors using a tradi—
`tional configuration, where the transistors are
`positioned horizontally creating a continuous
`row of transistors with the gates positioned verti—
`cally. Similar architectures can be found else—
`where [6, 8, 9,11,15,17,19—23, 26, 29—31 , 35].
`
`
`
`
`
`Fig. 4. Sea-of—gates microarchitecture. (a) Fishbone structure, 1A, 1B represent the transistor gates; (b) interleaved structure,
`1A, 1B, 1C, 2A, 2B, 2C represent the transistor gates; (c) octagon structure, 1A, 1B, 1C, 1D, 1E, 1F, 1G, 1H represent the
`transistor gates.
`
`436
`
`
`
`Microelectronics Journal, Vol. 26, No. 5
`
`New microarchitectures, however have recently
`appeared, presenting other advantages and new
`layout options for the sea-of-gates approach, for
`instance, the interleaved sea-of—gates image [37].
`This new architecture, S—MOS interleaved sea—
`of-gate’s architecture, presents a basic cell config-
`uration composed offlour two—layer metal CMOS
`gate arrays, in which the p- and n—channel source/
`drain regions and gate are laid out in parallel, as
`depicted in Fig. 4b. This parallelism facilitates the
`cell
`interconnections.
`that can be performed
`through straight lines over the cells without the
`need for changing metal direction, hence and so
`using only a single layer of metal. As compared
`with a conventional sea—of—gates basic cell archi—
`tecture, diffusion regions, both p- and n—channels,
`are slightly wider in the horizontal direction, and
`polysilicon gates 1A to 1C and 2A to 2C are not
`vertical in the column direction. The S-MOS
`
`interleaved architecture also claims to oEer a good
`answer for dense and. large circuits, with a high
`percentage of gate utilization, leaving at least 70%
`of the routing resources for global routing.
`
`Another powerful new sea-of-gates architecture
`is
`the octagon image [8]. This architecture
`presents a very regular and symmetrical structure
`in which the transistors are not arranged entirely
`
`in a single vertical or horizontal sequence, but
`symmetrically distributed as in an octagon. The
`cells in this structure can also be mirrored
`
`regarding the 45° mirror axis. Octagon image
`uses three levels of metal in a 0.8 pm CMOS
`process. As depicted in Fig. 4c, each basic cell is
`composed of four groups of four pairs of tran—
`sistors, symmetrically positioned. The gates are
`separated as a fishbone structure (1A to 1H).
`
`5.2 Macroarchitecture
`
`This is the higher level, in which the macrocells
`will be created as a combination of core cells. It
`
`is basically characterized by the number of core
`cells used to build the macro—architecture
`
`(macro-RAM block, multiplier, and so on), and
`the distribution of functions that deal with the
`
`distribution of the core cell on the master image
`(Fig. 5). The macroarchitecture can be further
`divided into three sub—groups:
`
`the
`o Uny’orm distribution — in this approach,
`distribution function is combined with one or
`
`more basic cells, so as to build all possible
`fimctional cells into the array (Fig. 5a).
`
`0 Channel distribution function - this style is simi—
`lar to the older gate array architecture,
`in
`
`D DD DDD DDDE
`
`l UD
`
`DDUDUUUDDDD
`
`DDDUDDDDDD
`DUDDDDDUDDUU DDDDDDUDDD
`DDDDDUDDUDDD
` DUDUDDDDDDU
`UDUDDDDUUDD
`UDDDDDDDDUD
`
`UDDDDDUDDDD
`
`DDDDDDDUDDD
`
`UDUDDDDDDD
`
`Fig. 5. Sea-of—gates macroarchitectures.
`
`(a) Uniform distribution function;
`distribution function.
`
`(b) channel distribution function;
`
`(c) block
`
`437
`
`
`
`M.E. de Lima and D.J. Kinniment/Sea-of—gates architecture
`
`which channels were defined to support
`routing. In general, this kind of architecture
`comprises either one or more cores to support
`the distribution functions (Fig. 5b).
`
`contained in a CAD system, and placed on a
`silicon slice rather than a PCB. Many other
`logic combinations can be provided to opti—
`mize the integration of the circuits.
`
`0 Block distribution function — still largely used. In
`this approach, the distribution functions are
`used almost solely as a combination of two or
`more core cells (Fig. 5c). The different core
`cells could be used to implement diEerent
`functions such as memory, analogue circuits,
`and so on.
`
`0 General cells — Also called parameterized cells
`(paracells) these cells, which usually involve a
`very regular structure, are generated by soft—
`ware in different sizes according to the need of
`the circuits. Examples of this class of cells are
`RAMs, ROMS, PLAs, multipliers, adders,
`and so on.
`
`Because of the use of more than one core cell
`
`the
`function(s),
`distribution
`different
`and
`applicability of micro- and macro—architecture
`becomes limited, directly influencing the flex—
`ibility of the placement of the functional cells on
`the array. Recent sea—of—gates architectures [29,
`31, 36, 37] have used just a single core cell to
`generate all
`the macro-architectures without
`routing channels. This approach allows
`the
`implementation of different
`design
`styles,
`supporting,
`for
`instance, dynamic and static
`libraries [35], and analogue circuits [23]
`in a
`homogeneous environment. The designer
`is
`now able to build up circuits which can share
`the regular structure, together with the placing
`of random logic blocks anywhere on the array
`[8, 22, 37, 38] without suffering substantial
`disadvantages.
`
`Taken together, these new architectures and the
`evolution of CAD tools has motivated a massive
`
`increase in the use of cell-based design using sea—
`of-gates. Currently, a designer can choose func—
`tional cells of varied complexity, which are pre-
`defined and pre—characterized in system libraries.
`These libraries can support different design
`styles, as follows:
`
`0 Standard cells (static or dynamic libraries) —~ We can
`compare these to the traditional CMOS
`families like 74C00. However, rather than
`selecting packaged devices from a catalogue,
`they are
`chosen fi'om software
`libraries
`
`438
`
`o Snpracells — also known as megacclls, these are
`larger unparameterized macrocells,
`such as
`microprocessors, A/D converters, and so on.
`
`Because of this wide range of possibilities in the
`physical layout of sea—of—gates, new tools have to
`be developed to deal with the extra complexity
`and high level of integration required. One of
`the most difficult problems for the layout of sea-
`of-gates is that there is no easy way to move the
`cells once placed in the array. This flexibility in
`placement can be very important, and heuristics
`should be developed to deal with the need to
`look ahead to later stages of the layout, such as
`global and detailed routing, considering free
`space available for routing, congestion, and so
`on, thus optimizing the silicon area available to
`guarantee the routability at every stage of the
`design.
`
`6. Conclusion
`
`important
`is one of the most
`Sea-of—gates
`microelectronic
`design
`styles
`to appear
`in
`recent years. Due to its flexibility and density,
`it
`is an ideal solution for highly—integrated
`systems that require a low and medium volume
`production. With the industry offering sea-of-
`gates arrays with over a million transistors, it is
`possible to implement complete systems or
`powerful microprocessors
`in a single array,
`with a density comparable in some cases to a
`full—custom style.
`
`
`
`Microelectronics Journal, Vol. 26, No. 5
`
`Acknowledgments
`
`This work was supported by the Federal Brazi—
`lian Agency for Post—Graduate Education,
`CAPES, Brazil.
`
`References
`
`[1]
`
`[2]
`
`M. Beuner, J.P. Kernhof and B. Hoefllinger, New
`directions in semi-custom arrays.J Solid-State Circuits,
`23(3) (Jun. 1988) 728—735.
`C. Piguet, E. Dijkstra, G. Berweiler, C. Voirol, M.
`Staugler and M. Joss, Module generators for sea-of—
`gates and fiill-custom ICs, Custom Integrated Circuits
`Cori, 1988, p. 349—652.
`[3] J. Adams, J. Van Dun, P. Guebis and M. Ratier,
`Technology Independent ASIC design methodology,
`Electr. Comm, 65(2) (1992) 168—181.
`HITACHI. HG625 Series. ASIC Devices Hitachi,
`1991.
`H. Kubasava, G. Goto, S. Tsutsumi, Y. Suehiro and
`T. Shirato, Layout
`to High—density Channelless
`masterslice, Custom Integrated Conf, 1987, p. 48~51.
`F. Anderson and J. Ford, A 0.5 micron 150k chan—
`nelless gate—array, [BEE CICC, 1987, pp. 35—38.
`M.E. De Lima and J. D. Kinniment, A force—directed
`placement algorithm with simultaneous global rout—
`ing for sea—of—gates architecture, VIII Congress
`(yr
`SBMICRO, Section III, September 1993, p.III.1—
`[11.7.
`
`[4]
`
`[5]
`
`[6]
`
`[7]
`
`[8]
`
`[9]
`
`[10]
`
`[11]
`
`[12]
`
`[13]
`
`[14]
`
`P. Groeneveld and P. Stravers, Ocean: the Sea-of-Cates
`Design System Manual, Delft University of Technol-
`ogy, Faculty of Electrical Engineering, 1993 (Tech-
`nical manual).
`A. Hui, A. Wong. C. Dell’oca, D. Wong and R.
`Szeto, A 4.1k gates double metal HCMOS sea—of-
`gates array,
`IEEE Custom Integrated Conf, 1985,
`pp. 15—17.
`T. Kobayashi, Latest techniques Enhance gate array
`integration, J. Electr. Eng. (1991) 26—29.
`Y. Kuramatsu, M. Ucda, T. Arakawa, M. Terai and
`S. Asai, A 540k-‘transistor CMOS variable—track
`Masterslice, IEEE J. Solid-State Circuits, SC-22 (2)
`(Apr. 1987) 192—201.
`T. Ohtanj, K. Yoshida, Y. Matsuda, M. Murayama
`and S. Toyoda, VLSI High—Speed CMOS SOG
`Family CMSO6, NBC Res. 8 Development J., 25
`(1989) 22—25.
`B.T. Preas and RC}. Karger, Placement, Assignment
`and Floorplanning, In: B. Lorenzetti, M.J. Lorenzetti
`(Eds), Physical Design Automation of VLSI Design, 1988,
`p. 87440.
`A. Sangiovanni—Vincentelli, Automatic layout of
`integrated Circuits,
`in: De Micheli, Sangiovanni-
`
`[15]
`
`[16]
`
`[17]
`
`[18]
`
`[19]
`
`[201
`
`[21]
`
`[22]
`
`[23]
`
`[24]
`
`[25]
`
`[26]
`
`[27]
`
`[28]
`
`[29]
`
`Vincentelli, A. Antogneti and P. NijhoE (Eds), Design
`Systems for VLSI Circuits, Martins Nijhoff, 1987,
`pp. 113—195.
`SHARP, L298 Series channel—free gate-array (Sea-
`of—gate), Sharp Corporation, 1990.
`Y. Suechiro, D. Miura, M. Naothoh, S. Tsutsumi
`and T. Shirato, A 120k gates usable sea-of—gates
`packing 1.3M transistors, Custom Integrated Circuits
`Conf., 1988, pp.519—522.
`Y. Suechiro and N. Matsumara, Development of sea—
`of—gates, Fujitsu Scientyic Tech. J., 24 (Dec. 1988) 318—
`327.
`
`P.J. Hicks, R.A. Cottrell and TA. York, Overview
`of semi custom IC design, Microproc. Microsyst. (1988)
`245—259.
`
`HJ. Veendrick, D.A.J.M. Van Den Elshout, DW.
`Harberts and T. Brand, An Eflicient and Flexible
`Architecture for High-density Gate Arrays, J. Solid-
`State Circuits, 25(5) (Oct. 1990) 1153—1157.
`F. Anderson and J. Ford, A 150k Channelless gate
`array design in 0.5 CMOS technology, J.Solid-State
`Circuits, 23(2) (Apr. 1988) 520—522.
`M. Beuner, J.P. Kernhof and B. Hoefilinger, The
`CMOS gate forest: An efficient and flexible high-
`performance ASIC design environment, J Solid—State
`Circuits, 23(2) (Apr. 1988) 387—398.
`R. Blumberg and C. Waggoner, An 840k transistor
`Sea—of—Gates 1.2um HCMOS technology,
`IEEE
`ISCC, 1988, p. 74—75.
`P. Duchene and MJ. Decercq, A High Flexible sea-
`of—gates Structure for Digital and Analog Applica-
`tions, J Solid-State Circuits, 24(3) (Jan. 1989) 578—
`584.
`
`C.P. Hsu, R.A. Perry, S.C. Evans, J. Tang and J.Y.
`Liu, Automatic Layout of Channelless gate array,
`Custom Integrated Circuits Conf., 1986, pp. 281—284.
`E.S. Kuh and T. Ohgtsuki, Recent advances in
`layout. Proc. IEEE, 78(2) (Feb. 1990) 237—263.
`F. Murabayashi, Y. Nishio, H. Maejima, A. Wata—
`nabe, S. Shukuri, T. Nishida and K. Shimohigashi, A
`0.5um BiCMOS Channelless gate-array, Custom
`Integrated Circuits Confi, 1989, pp. 8.7.1—8.7.4.
`I. Ohkura, T. Noguchi, K. Sakashita, H. Ishida, T.
`Ichiyama and T. Enomoto, Gate Isolation — A
`novel
`basic
`cell
`configuration,
`ICAD,
`1982,
`pp.307—310.
`W.A.M. Van Voije and GJ. Declerck, Advanced
`CMOS Gate Array Architecture Combining Gate
`Isolation and Programmable Routing Channels, J.
`Solid-State Circuits, SC—20 (2) (1985) 469—480.
`M. Okabe, Y. Okuno, T. Arakawa, I. Tomioka, T.
`Ohno, T. Noda, M. Hatanaka and Y. Kuramitsu, A
`400k—transistor CMOS Sea—of—gates
`array with
`continuous track allocation, J. Solid—State Circuits,
`24(5) (Oct. 1989) 1280-1285.
`
`439
`
`
`
`M.E. de Lima and D.J. Kinniment/Sea-of-gates architecture
`
`[30]
`
`[31]
`
`[32]
`
`[33]
`
`Y. Okuno, M. Okabe, T. Arakuwa, T. Tomioka, T.
`Ohno, T. Noda and Y. Karamitsu, 0.8 ,um 1.4 Mtr.
`CMOS SOG based on column Macro-cell, Custom
`Integrated Circuits Conf, 1989, pp. 82.1—82.4.
`M. Okabe, Y. Okuno, T. Arakawa, T. Tomoika, T.
`Ohno, T. Noda, M. Hammaka and Y. Kuramistu, A
`CMOS sea-of—gates array with continuous tracks
`allocation, ISSCC-89, 1989, pp. 180—181.
`B. Korke, HJ. Promel and A. Steger, Combining
`Partitioning and Global Routing in Sea—of—Cells
`Design, ICCAD, 1989, pp. 98—101.
`C-M. Kyung, P.V. Kraus and D. A. Mlynski, Adap—
`tive cluster growth: a new algorithm for ciruit place—
`ment
`in rectilinear regions, Computer-Aided—Design,
`24(1) (1992) 27—35.
`
`[34] E. Shragotwitz, J. Lee and S. Shani, Algorithms for
`physical design of sea—of-gates chips, Computer—Aided
`Design, 20(7) (Sept. 1988) 382—397.
`[35] ]. Kemofl‘, M. Seller, M. Buender, B. Hoelfllinger, B.
`Laquai, and I. Scheindler, Mixed static and domino
`Logic CMOS gates forest, Solid—State Circuits, 25 (Apr.
`1990) 396-402.
`B. Egan, More from less is promise of new sea—of—
`gates architecture, Computer Design (Jan. 1992) 28~
`29.
`
`[36]
`
`[37]
`
`[38]
`
`B. Egan, Oki enhances sea—of—gates performance,
`Computer Design (Jun. 1992) 127.
`in
`S. Hurst, Custom microelectronics techniques,
`Custom VLSI Microelectronics, Prentice-Hall, 1992,
`pp. 129*157.
`
`440
`
`